Proto commits in MystenLabs/sui

These commits are when the Protocol Buffers files have changed: (only the last 100 relevant commits are shown)

Commit:728e442
Author:Nick Vikeras
Committer:GitHub

rpc: require public positions on transaction and event cursors, and drop their checkpoint label and the legacy cursor format (#28541) ## Description #28527 (in 1.83) gave every transaction and event ledger cursor its public position and bounds scans by `tx_seq` alone. That leaves three parts of the cursor format with no job, and this removes them for 1.84: - **The checkpoint label** (tag 1 of `TransactionsPosition`/`EventsPosition`, now reserved). - **Transaction and event cursors without a public position.** They fail with `INVALID_ARGUMENT`, so every `List*` frame carries `cursor_position`. - **The decode fallback for the original cursor format** (tags 1–4 of `CursorToken`). Those cursors fail with `INVALID_ARGUMENT` too. Only the first and fourth of the sixteen commits change behavior; the rest are renames, the crate fold, simplifications and tests. ## Mini design doc ### Design A cursor is an opaque `tx_seq` (plus `event_index` for events), which bounds the scan, and a public position (`checkpoint`, `transaction_index`, `event_index`), which only labels it. With the label gone, `ResolvedScan` tracks the public position of where a scan ends (`end_public_position`), so a page's terminal cursor and its frame's `cursor_position` are known without a read. A scan frontier is still placed by the one index row the scan already reads (`RowPlacement`). The only frontier no row places is the ascending genesis frontier, which is placed at genesis; the frontier cases that existed only for the label, and the `Internal` errors that went with them, are gone. ### Decisions - **Keep `tx_seq` as the scan bound.** Rejected: bounding scans by the public position. It makes reads no cheaper (slightly more expensive on the fullnode), needs new handling for cursors below the pruned floor, and 1.83 rejects cursors without `tx_seq`, so 1.84 has to keep writing it anyway. Also rejected: dropping the public position from cursors and placing each page end with a row read. That adds a read per page (a Bigtable round trip on kv), and a mixed 1.83/1.84 fleet would emit frames without `cursor_position`, which GraphQL rejects. - **Reject position-less cursors instead of keeping a fallback.** Only servers before 1.83 mint them, and the fallback was what needed the label and its edge cases. A client holding one restarts the scan from a ledger position. - **Track the terminal cursor's public position in `ResolvedScan`, not derive it when rendering.** A derivation (serving floor, request cursor or window edge) has no complete fallback: it would need an `expect` in `sui-node`, which builds with `panic=abort`, or an invented default. - **Keep `end_public_position` and `end_cursor_public_position()` separate.** The method applies the request-cursor override on every read. Storing it once in `resolve` would be equivalent today, but only because no code that moves the end after `resolve` (`apply_serving_floor`, and `list_checkpoints`' filtered-window floor) can land on a request cursor's position, a rule that spans three files. - **A transaction cursor's public position has no `event_index`**, dropped on both decode and encode. A client can forge one; it would only change the client's own terminal cursor and `cursor_position`, but nothing needs it. - **Fold `sui-rpc-cursor` into `sui-rpc-api` as `ledger_history::cursor`.** It was a crate so GraphQL could share the format; GraphQL has carried ledger cursors as opaque bytes since #28516. - **Keep the old tags reserved** (`reserved 1 to 4` on `CursorToken`, `reserved 1` on the position messages). Clients and GraphQL cursors hold 1.83 cursors indefinitely, and 1.83 servers write tag 1 during the rollout. ### Surprises - A transaction cursor's forged public `event_index` passed straight into the terminal cursors minted from it. The fourth commit drops it on decode, and the twelfth on encode. - A 1.83 server echoes a position-less request cursor back unchanged as its terminal cursor, so such cursors can come from 1.83 servers too. The release note says so. - Once the label went, an unplaced frontier (only ever the ascending genesis frontier) behaved exactly like checkpoint 0 everywhere, and a placed frontier's separate checkpoint always equaled its point's, so `PlacedFrontier` became a plain `LedgerPoint` on both backends, with the genesis frontier placed at `LedgerPoint::GENESIS`. ### Impact and rollout - **Depends on #28527 being in 1.83** (#28540, merged into `releases/sui-v1.83.0-release`). With it, a 1.84 server rejects only cursors that servers before 1.83 minted. - **Mixed 1.83/1.84 fleet:** 1.84 decodes 1.83 cursors and skips the label. 1.83 decodes 1.84 cursors: it reads the public position at tag 9, reads the missing label as 0, and only copies the label into cursors it mints. The position fields stay `optional`, so a genesis `tx_seq` 0 stays on the wire, which 1.83's decoder requires. - **Clients:** pre-1.83 cursors and original-format cursors fail with `INVALID_ARGUMENT`. Cursor bytes change (no label), so the GraphQL snapshot cursors are regenerated; nothing else in the snapshots changes. - **No protocol-config gating:** this is RPC only. - **Follow-ups:** MystenLabs/sui-rust-sdk#342 removes the SDK's handling of frames without `cursor_position` once 1.84 is on mainnet. GraphQL's check for frames without `cursor_position` stays while it can be served by 1.83. ### Test coverage - **Covered:** cursor encode/decode unit tests (1.83 bytes with the label, position-less and original-format rejection, `event_index` dropped both ways). The model sweeps in `query_options` and kv `resume_tests` check scan bounds, end positions and `cursor_position` against a model for every cursor kind and ordering. e2e `list_ledger_history` and `kv_rpc_tests` run real servers: cursors sent back in the opposite direction, past-tip and pruned-floor requests. - **Could slip:** no test runs a 1.83 server against 1.84 cursors; that direction is argued from the 1.83 decoder, while the reverse is pinned by 1.83-shaped bytes. The e2e tests ran under nextest, not `cargo simtest`. ## Test plan - `ledger_history::cursor`: a transaction or event cursor carrying the label, as 1.83 mints it, decodes and re-encodes without it. Transaction and event cursors without a public position, and cursors in the tag 1–4 schema, are rejected. A transaction cursor whose public position carries an event index decodes and encodes without it (`transaction_cursors_drop_a_public_event_index`). - e2e: the fullnode answers a transaction cursor without a public position with `INVALID_ARGUMENT`. - `test_list_events_frontier_below_the_pruned_floor_is_out_of_range` also sends its frontier with a public position forged to the floor checkpoint and expects `OutOfRange`: the floor check reads the cursor's scan position, not its public position. - kv `scan_limit_terminal_frames_use_transaction_domain_in_both_directions` gains rows where the frontier advances the covered checkpoint. - Removed: label expectations, cases that sent cursors without a public position, frontier tests that guarded the label, and rows for unplaced frontiers other than ascending genesis. One e2e case changed: a past-tip `after` cursor, which now carries a public position, gets that point as its frame's `cursor_position` instead of none. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [x] gRPC: `ListTransactions` and `ListEvents` reject cursors minted by servers older than 1.83, including such cursors a 1.83 server returned, and every `List*` method rejects cursors in the original cursor format, with `INVALID_ARGUMENT`; restart such a scan from a ledger position instead. Every `List*` frame's `Watermark` carries `cursor_position`. - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:bb2bd9a
Author:Nick
Committer:Nick

rpc: require public positions on transaction and event cursors, and drop their checkpoint label and the legacy cursor format

Commit:f5727ef
Author:Nick
Committer:Nick

rpc: require public positions on transaction and event cursors, and drop their checkpoint label and the legacy cursor format

Commit:f8a0138
Author:Nick
Committer:Nick

rpc: require public positions on transaction and event cursors, and drop their checkpoint label and the legacy cursor format

Commit:022b96f
Author:Nick Vikeras
Committer:GitHub

[1.83] Cherry-pick RPC, GraphQL, and kv-rpc ledger and signature-verification changes (#28540) ## Description Cherry-picks the following RPC, GraphQL, and kv-rpc changes from `main` to the 1.83 release branch, in `main` order, via `git cherry-pick -x`: - #28395 graphql: decouple signature verification from protocol activation - #28470 graphql: opt-in address-alias checking for verifySignature - #28412 grpc: verify both supported zkLogin circuits - #28471 grpc: opt-in address-alias checking for VerifySignature - #28483 graphql: end empty List API pages between two cursors - #27666 sui-indexer-alt-graphql: remove Postgres pagination for transactions and events - #27712 sui-kv-rpc: remove enable-list-apis flag, always serve the List APIs - #27723 sui-indexer-alt-reader: merge AlphaLedgerGrpcReader into LedgerGrpcReader - #28516 (rebase-merged as 5 commits): - rpc: start and end positions within a checkpoint for ListTransactions and ListEvents - [graphql] Own transaction, event, and checkpoint cursors; carry ledger cursors as opaque bytes - rpc: mark ledger OUT_OF_RANGE errors as above latest or below earliest available - rpc: end an ascending page at its after cursor past the tip instead of failing OUT_OF_RANGE, and read light-client events through the ledger streams - Bump sui-rust-sdk to fc8391c - #28534 [graphql] Include address-balance changes when matching affectedAddress in live subscriptions - #28527 rpc: report each ledger watermark's resume point as a public cursor_position - #28535 rpc: match the sender in AffectedAddress transaction filters - #28539 sui-kv-rpc: fix resume_tests call broken by concurrent merges of #28527 and #28535 (open on `main`; picked from its head commit `3ce33fc45ca`) #28412 is a prerequisite of #28471, and #28483, #27666, and #27723 are prerequisites of #28516: without them those picks conflict. #28395 is the GraphQL counterpart of #28412. All picks applied without conflicts. The `sui-rust-sdk` bump (`1313d51` → `8c356d6`) only changes the `sui-rpc` crate (vendored protos and client ledger streams); `sui-crypto` and `sui-sdk-types` are unchanged. ## Test plan `cargo check --locked --workspace --all-targets` passes on this branch. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework: --------- Co-authored-by: Ashok Menon <ashok@mystenlabs.com> Co-authored-by: hayes-mysten <135670682+hayes-mysten@users.noreply.github.com> Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com> Co-authored-by: Evan Wall <evan.wall@mystenlabs.com>

Commit:b2fc658
Author:Nick Vikeras
Committer:GitHub

rpc: report each ledger watermark's resume point as a public cursor_position (#28527) ## Description Clients can't order ledger cursors: `Watermark.cursor` is opaque, so the Rust SDK's ledger streams and GraphQL's subscription backfill compare cursor bytes and guess frame order checkpoint by checkpoint. MystenLabs/sui-apis#48 adds `Watermark.cursor_position`, the resume point as a public `(checkpoint, transaction_index, event_index)`. This PR has the fullnode and kv-rpc `List*` APIs and the fullnode `SubscribeTransactions`/`SubscribeEvents` set it on every frame. Apart from the two error fixes below, no request changes outcome, and no storage reads are added. - Item frames take the position from the item: past it when ascending, at it when descending. - Frames without an item place the cursor's `tx_seq` using the `tx_seq_digest` row the scan already reads. - The internal cursor (`sui.rpc.cursor.v1.CursorToken`) also carries the public position of its coordinate, so a server can place any cursor it minted without a lookup: a page ending at a request cursor, or the past-tip echo of an `after` cursor. Scans still bound by `tx_seq`, so this position is only reported back, never trusted for bounds. Old servers skip the new field. - The position is left unset only on a final frame named by a request cursor minted before this change, when the request has no matching start/end fields. - `ABOVE_LATEST_AVAILABLE` for a descending cursor past the tip reports the checkpoint of the cursor's public position as `requestedCheckpoint`. Checkpoint cursors and cursors minted before this change report their internal checkpoint label, as before. That was the label's only public output, so a later release can stop writing it for transaction and event cursors. - Cursor bytes gain the new field. Old SDKs and old GraphQL compare cursor bytes, so while fullnodes and kv-rpc run mixed versions they see two encodings of one point and take their replay paths more often. - `sui_kvstore::BigTableClient::resolve_tx_checkpoints` lost its last caller and is deleted. - Bumps sui-rust-sdk to the vendoring commit from MystenLabs/sui-rust-sdk#336. Re-pin to the merged master commit once #336 and sui-apis#48 merge. Error fixes from end-to-end testing: - Fullnode: a descending request from a cursor past the tip down to a `start_checkpoint` below the pruning floor can never be served, but failed with the retryable `ABOVE_LATEST_AVAILABLE`, which the Rust SDK retries until the node catches up (forever for a forged cursor). It now fails with `BELOW_EARLIEST_AVAILABLE`. - kv-rpc: an invalid filter returned an empty `LEDGER_TIP` page when the request left nothing to scan (a window or `after` cursor past the tip). The filter is now compiled first, so it is `INVALID_ARGUMENT`, as on the fullnode. **1.83:** this builds on b3edeaf07a3 (#46 start/end positions), 45735305f99 (`ABOVE_LATEST_AVAILABLE` reasons) and 7425bd74e7d (ascending past-tip echo). None of them is on `releases/sui-v1.83.0-release` yet, so a 1.83 cherry-pick needs all of them. Related PRs: - Proto: MystenLabs/sui-apis#48 - SDK vendoring: MystenLabs/sui-rust-sdk#336 - SDK resumption (merge after this ships): MystenLabs/sui-rust-sdk#337 - GraphQL (stacked on #28515): #28528 ## Test plan New tests, on both backends: - Every frame's position splits the delivered items correctly and never regresses. - Paging by position plus cursor returns exactly what one unbounded scan does, in both directions. - Past-tip, mid-checkpoint and scan-limit terminals, and Subscribe item and progress frames. - kv `resume_tests.rs` and the e2e `list_ledger_history`/`subscription_service` tests cover the same through the wire. - Cursors round-trip their position, cursors minted before this change still decode and bound scans exactly as before, and echoed cursors are byte-identical to the request's. - Descending from a past-tip `before` down to a start below the floor is `BELOW_EARLIEST_AVAILABLE` in every fullnode List API; starting at the floor keeps `ABOVE_LATEST_AVAILABLE` (e2e). An unparseable filter is `INVALID_ARGUMENT` on each kv-rpc List API with a window or `after` past the tip. - A descending cursor past the tip reports its public position's checkpoint as `requestedCheckpoint`, and its label when it has no public position. - Crossing `after`/`before` cursors below the pruned floor still return an empty `CURSOR_BOUND` page (e2e); with a cursor minted before this change and no position fields, its position is unset. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [x] gRPC: `ListCheckpoints`, `ListTransactions`, `ListEvents`, `SubscribeTransactions` and `SubscribeEvents` frames carry `Watermark.cursor_position`, the frame's resume point as a public ledger position. Use it to order frames and drop duplicates instead of comparing cursor bytes. kv-rpc `List*` requests with an invalid filter fail with `INVALID_ARGUMENT` even when nothing is left to scan. On a fullnode, a descending request whose `start_checkpoint` is below the pruning floor fails with `BELOW_EARLIEST_AVAILABLE` even when its `before` cursor is past the tip. - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:aea6c03
Author:Nick
Committer:Nick

rpc: report each ledger watermark's resume point as a public cursor_position

Commit:015dfbe
Author:Nick
Committer:Nick

rpc: report each ledger watermark's resume point as a public cursor_position

Commit:17beead
Author:Nick
Committer:Nick

ledger-history: drop the checkpoint label from transaction and event cursors

Commit:b87d951
Author:Nick
Committer:Nick

ledger-history: drop the checkpoint label from transaction and event cursors

Commit:eb8dc66
Author:Nick Vikeras
Committer:GitHub

[1.83] Cherry-pick kvstore, kv-rpc, and indexer changes (#28465) ## Description Cherry-picks the following kvstore / kv-rpc / indexer changes from `main` to the 1.83 release branch (in order, via `git cherry-pick -x`): - #28426 ledger-history: end pages resumed past their window with the window's reason - #28414 [kvstore] Add opt-in object tombstones to the objects table - #28443 rpc-store: write each checkpoint as soon as it is next in order - #28444 indexer-alt-framework, kvstore: write batches and watermarks as soon as they are ready - #28441 [kvstore] Detect and route around stalled BigTable channels - #28447 [kvstore] Add an opt-in gRPC server with a catch-up readiness health check - #28451 [kv-rpc] Optionally set the served checkpoint from kvstore indexers' watermark streams #28451 was not yet merged to `main` when this branch was created; the pick is of its head commit `c55e29b30db`. All picks applied without conflicts. ## Test plan `cargo check --locked --all-targets` on every touched crate (`sui-kvstore`, `sui-kv-rpc`, `sui-rpc-store`, `sui-rpc-api`, `sui-indexer-alt-framework`, `sui-indexer-alt`, `sui-indexer-alt-consistent-store`, `sui-indexer-alt-e2e-tests`, `sui-analytics-indexer`, `sui-checkpoint-blob-indexer`). ## Release notes - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:c52e290
Author:Nick Vikeras
Committer:GitHub

[kvstore] Add an opt-in gRPC server with a catch-up readiness health check (#28447) ## Description Kubernetes marks a new kvstore pod ready about 18 s after it starts, while it is still replaying minutes of history. A rollout can therefore replace every writer with one that is behind: on 2026-10-20 that left the data in Bigtable 96 s behind the chain. With `--rpc-listen-address`, kvstore serves gRPC health checks that stay NOT_SERVING until every pipeline is within 1 s of its peers (`--peer-url`, the other writers) and of the watermarks stored at startup. A readiness probe on that port keeps the old pod writing until its replacement has caught up. Without the flag, nothing changes. Notes: - An instance restarted during a chain halt stays not ready until the chain resumes, and the instances it would replace keep serving. - With no reachable peer (a single instance, or the first rollout onto this change), an instance compares only with the watermarks stored at startup, so it can become ready while still behind by about as long as its catch-up took. ## Test plan Unit tests cover the readiness rules, peer following, and server shutdown. On mainnet, ran this change (cherry-picked onto 1.82.2) on the three `mainnet-workloads` writers and restarted all three at once, 20 times in a row. Indexer lag below is how far the data in Bigtable (the stalest pipeline, from the freshest writer) trails the median Mysten validator, from Prometheus. | | Indexer lag, max | |---|---| | Steady state, 2 hours | 0.24 s | | Last 3 rollouts without this change | 96 s, 0.40 s, 0.24 s | | 20 rollouts with this change | 0.40 s | Without the gate, lag depended on whether some writer happened to stay current. With it, new pods were up to 572 s behind while catching up, but from the second rollout on none was marked ready until it was within 1 s of its peers, so the old pods kept writing throughout. (The first rollout had no peers to compare with; see the notes above.) No pod restarted, and every rollout finished in 67–176 s. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:7ca87a9
Author:Nick
Committer:Nick

ledger-history: drop the checkpoint label from transaction and event cursors

Commit:e9d56e6
Author:Nick
Committer:Nick

[kvstore] Add an opt-in gRPC server with a catch-up readiness health check With --rpc-listen-address, sui-kvstore-alt serves gRPC with grpc.health.v1, reflection, and sui.kvstore.v1alpha.WatermarkService. The overall health status and every registered service report NOT_SERVING until this instance has caught up after startup, then latch to SERVING, so a Kubernetes readiness probe holds a rollout until the new pod has caught up. Without the flag, the indexer binds no port, reads no stored watermarks at startup, and does not track readiness; --peer-url and --readiness-max-lag-ms require it. Readiness: - DurableProgress records, per pipeline, the committer watermark through which data rows are durable. Concurrent and sequential pipelines record before the watermark write, and bitmap pipelines when the generation worker promotes, so a failed or lost watermark write does not delay it. - An instance is caught up when every pipeline is within --readiness-max-lag-ms (default 1s of chain time) of the stored watermark row read at startup, and of the durable watermarks its peers (--peer-url) stream. Each pipeline must commit once after startup, so a restart during a chain halt waits for new checkpoints. - Peers hold up readiness for at most as long again as catching up to the startup rows took (at least 60s), so an instance that stays behind its peers still becomes ready. Unreachable peers do not hold it up, and a pipeline whose peer watermark trails the peers' most advanced watermark by more than 60s of chain time is backfilling and is compared only with its startup row. WatermarkService streams, per pipeline, the higher of the stored watermark row read at startup and DurableProgress: every pipeline on the first response, then each advance. Starting from the stored rows gives a subscriber the height of every pipeline in its first response, including a pipeline this instance has not advanced since it started (for example during a chain halt); rows that readers treat as missing (no checkpoint, or below reader_lo) are left out. Streams read a copy that a relay task keeps up to date, so a committer's update wakes only the relay rather than every open stream on the commit path. Peer followers keep the highest watermark per pipeline, resubscribe after a stream ends or fails, and use HTTP/2 keepalive to leave a peer that stops responding; the server's keepalive closes unresponsive subscribers, so graceful shutdown does not wait on them. Shutdown clears every health status, which ends open Health/Watch streams. get_watermark_for_pipelines is built on get_pipeline_watermarks, one batched read-and-decode path for watermark rows.

Commit:9ee36ba
Author:Nick
Committer:Nick

[kvstore] Add an opt-in gRPC server with a catch-up readiness health check With --rpc-listen-address, sui-kvstore-alt serves gRPC with grpc.health.v1, reflection, and sui.kvstore.v1alpha.WatermarkService. The overall health status and every registered service report NOT_SERVING until this instance has caught up after startup, then latch to SERVING, so a Kubernetes readiness probe holds a rollout until the new pod has caught up. Without the flag, the indexer binds no port, reads no stored watermarks at startup, and does not track readiness; --peer-url and --readiness-max-lag-ms require it. Readiness: - DurableProgress records, per pipeline, the committer watermark through which data rows are durable. Concurrent and sequential pipelines record before the watermark write, and bitmap pipelines when the generation worker promotes, so a failed or lost watermark write does not delay it. - An instance is caught up when every pipeline is within --readiness-max-lag-ms (default 1s of chain time) of the stored watermark row read at startup, and of the durable watermarks its peers (--peer-url) stream. Each pipeline must commit once after startup, so a restart during a chain halt waits for new checkpoints. - Peers hold up readiness for at most as long again as catching up to the startup rows took (at least 60s), so an instance that stays behind its peers still becomes ready. Unreachable peers do not hold it up, and a pipeline whose peer watermark trails the peers' most advanced watermark by more than 60s of chain time is backfilling and is compared only with its startup row. WatermarkService streams DurableProgress to subscribers: every pipeline on the first response, then each advance. Peer followers keep the highest watermark per pipeline, resubscribe after a stream ends or fails, and use HTTP/2 keepalive to leave a peer that stops responding; the server's keepalive closes unresponsive subscribers, so graceful shutdown does not wait on them. Shutdown clears every health status, which ends open Health/Watch streams. get_watermark_for_pipelines is built on get_pipeline_watermarks, one batched read-and-decode path for watermark rows.

Commit:286c12a
Author:Nick

[kvstore] Add an opt-in gRPC server with a catch-up readiness health check With --rpc-listen-address, sui-kvstore-alt serves gRPC with grpc.health.v1, reflection, and sui.kvstore.v1alpha.WatermarkService. The overall health status and every registered service report NOT_SERVING until this instance has caught up after startup, then latch to SERVING, so a Kubernetes readiness probe holds a rollout until the new pod has caught up. Without the flag, the indexer binds no port, reads no stored watermarks at startup, and does not track readiness; --peer-url and --readiness-max-lag-ms require it. Readiness: - DurableProgress records, per pipeline, the committer watermark through which data rows are durable. Concurrent and sequential pipelines record before the watermark write, and bitmap pipelines when the generation worker promotes, so a failed or lost watermark write does not delay it. - An instance is caught up when every pipeline is within --readiness-max-lag-ms (default 1s of chain time) of the stored watermark row read at startup, and of the durable watermarks its peers (--peer-url) stream. Each pipeline must commit once after startup, so a restart during a chain halt waits for new checkpoints. - Peers hold up readiness for at most as long again as catching up to the startup rows took (at least 60s), so an instance that stays behind its peers still becomes ready. Unreachable peers do not hold it up, and a pipeline whose peer watermark trails the peers' most advanced watermark by more than 60s of chain time is backfilling and is compared only with its startup row. WatermarkService streams DurableProgress to subscribers: every pipeline on the first response, then each advance. Peer followers keep the highest watermark per pipeline, resubscribe after a stream ends or fails, and use HTTP/2 keepalive to leave a peer that stops responding; the server's keepalive closes unresponsive subscribers, so graceful shutdown does not wait on them. Shutdown clears every health status, which ends open Health/Watch streams. get_watermark_for_pipelines is built on get_pipeline_watermarks, one batched read-and-decode path for watermark rows.

Commit:9c67d9e
Author:Nick Vikeras
Committer:GitHub

ledger-history: make List cursors work in either direction (#28391) ## Description A `ListTransactions`/`ListEvents` cursor sent back in the opposite ordering (an `after` cursor reused as `before`, or the reverse) could skip items. Servers narrowed the scan by the cursor's checkpoint field, and the value of that field depends on the ordering of the scan that minted the cursor. Servers now bound scans by cursor positions only. The checkpoint field becomes a label that servers copy into the cursors they return. Below, L is the server's latest indexed checkpoint. Client-visible behavior on both the fullnode and kv-rpc: - A descending page bounded by `after: Item(X)` closes at the boundary after X, so continuing downward returns X. - A page whose starting cursor sits at the window's edge ends with the window's reason: `CheckpointBound`, or `LedgerTip` for an ascending scan that reaches L. It returns the same closing cursor and coverage claim as the page that issued the cursor. - A descending `after` cursor past L returns `OutOfRange`, as a starting cursor past L already does. - A descending request whose window reaches past L returns `OutOfRange`: an `end_checkpoint` above L + 1, or a `start_checkpoint` of L + 1 or more with `end_checkpoint` unset. Answering it would skip checkpoints the server has not indexed yet, because a descending scan starts at its window's high end. An ascending request past L still ends at L with `LedgerTip`, because the client resumes from there once the server catches up. - `Watermark.checkpoint` does not depend on the cursor's checkpoint field. A scan claims each checkpoint whose end it reaches, and its starting cursor vouches for the positions before the cursor. A page that resumes at the window's edge claims the window's last checkpoint (ascending) or first checkpoint (descending). A window with no checkpoints claims nothing. - On a pruning fullnode, a request that would read data below the pruning floor ends with `OutOfRange` ("requested data below earliest available"): - a `start_checkpoint` or `after` cursor below the floor, in either ordering; - a request with no `start_checkpoint` and no `after` cursor whose `end_checkpoint` is at or below the floor, or whose `before` cursor is at or below the floor; - a descending request with no `start_checkpoint` and no `after` cursor that reaches the floor: it streams everything down to the floor, sends a watermark frame at the floor that claims the floor checkpoint, and then fails with `OutOfRange`. The watermark's cursor resumes the walk on a server with more history. A request whose cursors leave nothing to scan reads no data, so it gets its usual empty page even when its window or cursor lies below the floor. For example, a descending `before` cursor at genesis with no `start_checkpoint` ends with `CheckpointBound` and claims checkpoint 0, which the cursor already vouches for. An ascending request with no `start_checkpoint` and no `after` cursor still starts at the floor. GraphQL sends `last:` pages with `end_checkpoint` one past its `checkpoint_viewed_at`. kv-rpc holds such a request until it reaches GraphQL's checkpoint floor (`x-sui-min-checkpoint`), so GraphQL over kv-rpc does not see the descending `OutOfRange`. The fullnode ignores that header, so GraphQL over load-balanced fullnodes gets `OutOfRange` on a `last:` page sent to a fullnode behind GraphQL's watermark, where before it got a page missing the newest items. Follow-ups: - GraphQL subscriptions read the cursor's checkpoint label; a separate PR moves them to `Watermark.checkpoint` and the item payloads. - A staged PR removes the label from transaction and event cursors. - The proto docs in sui-apis need updates. The `Watermark.checkpoint` doc still says a scan resumed mid-checkpoint leaves the field unset until it covers that checkpoint. The `start_checkpoint` doc says it defaults to genesis, but on a pruning fullnode an unset start begins at the lowest available checkpoint (`x-sui-lowest-available-checkpoint`). ## Test plan - Property tests over a model ledger (every cursor each kind of scan can mint, in every window and ordering, including windows that reach past L) check the admitted positions, the entry checkpoint, the end reason, and which requests return `OutOfRange`. They also check that each closing cursor resumes exactly the positions its far bound excluded. - Fullnode and kv e2e tests page real ledgers and send every returned cursor back in the opposite ordering. They also check that a page resumed at the window's edge returns the same end reason, closing cursor, and checkpoint claim as the page that issued the cursor, and that a descending window past L returns `OutOfRange`. - Fullnode e2e tests check that each floor case returns `OutOfRange` in all three List APIs: a cursor below the floor, a window ending at or below it with no lower bound, and a descending walk that reaches it. For the walk they check the items down to the floor, then the watermark frame claiming the floor checkpoint, then the error; resuming from that watermark's cursor in either ordering also returns `OutOfRange`. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [x] gRPC: `ListTransactions` and `ListEvents` cursors can be sent back in either ordering without skipping items. A page that resumes at its window's edge ends with the window's end reason and claims the window's edge checkpoint. These requests return `OutOfRange` instead of an empty or clamped result: a descending `after` cursor past the latest indexed checkpoint; and a descending `ListTransactions`, `ListEvents`, or `ListCheckpoints` request whose `end_checkpoint` is above the latest indexed checkpoint + 1, or whose `start_checkpoint` is above the latest indexed checkpoint with no `end_checkpoint`. On a pruning fullnode, every request that needs data below the pruning floor returns `OutOfRange`. A descending request with no `start_checkpoint` streams everything down to the floor, then a watermark at the floor, before the error. - [ ] JSON-RPC: - [x] GraphQL: Transaction and event connections served by the List APIs derive `hasNextPage`/`hasPreviousPage` from the end reason. A page that resumes at its window's edge, or at the latest indexed checkpoint from a cursor other than that checkpoint's own boundary, reports no further page where it reported one before. A `last`/`after` page whose `after` cursor sits exactly at the latest indexed checkpoint reports `hasPreviousPage: true` where it reported `false` before. - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:0b9a909
Author:Nick
Committer:Nick

ledger-history: drop the checkpoint label from transaction and event cursors

Commit:f3ec59d
Author:Nick
Committer:Nick

ledger-history: make List cursors work in either direction

Commit:fa78e84
Author:Nick

ledger-history: drop the checkpoint label from transaction and event cursors

Commit:3900326
Author:Nick
Committer:Nick

ledger-history: make List cursors work in either direction

Commit:27a1c7b
Author:stefan-mysten

[sui-fork] Seed objects, addresses, and dynamic fields into a running fork over gRPC

Commit:dc3e704
Author:Will Yang

[rpc] Package cursor lane for MovePackageService.ListPackages Adds the wire PackagesPosition variant ordered by (tx_seq, write_index) and admits it into the existing IntraTxCoordinate scan space — packages are the second per-transaction-indexed lane, so the coordinate, bounds lattice, resolution, and cursor arms are all reused rather than minted: one decode arm and a lane-scoped QueryOptions constructor is the whole rpc-api surface.

Commit:2234d9a
Author:Will Yang
Committer:Will Yang

[rpc] Package-write scan machinery for MovePackageService.ListPackages Groundwork for serving ListPackages (sui-rust-sdk#pending) from ledger history: - sui-rpc is pinned to the sui-rust-sdk commit carrying the ListPackages proto; re-pin to the merge commit once that PR lands. - sui-rpc-cursor grows Position::Packages { checkpoint, tx_seq, write_index }: package writes are sub-transaction items, mirroring Position::Events, with the same Item/Boundary resume inclusivity. - sui-rpc-api's ledger_history gains the package analogues of the event scan machinery: PackagePosition / PackageScanBounds / ResolvedPackageRange and QueryOptions::apply_package_cursor_bounds. The wire only addresses transactions, so mid-transaction cursors re-include their boundary transaction via PackageScanBounds::tx_range and the already-served writes are skipped during expansion — the same shape as the unfiltered ListEvents path.

Commit:27d390a
Author:Will Yang

[rpc] Package-write scan machinery for MovePackageService.ListPackages Groundwork for serving ListPackages (sui-rust-sdk#pending) from ledger history: - sui-rpc-cursor grows Position::Packages { checkpoint, tx_seq, write_index }: package writes are sub-transaction items, mirroring Position::Events, with the same Item/Boundary resume inclusivity. - sui-rpc-api's ledger_history gains the package analogues of the event scan machinery: PackagePosition / PackageScanBounds / ResolvedPackageRange and QueryOptions::apply_package_cursor_bounds. The wire only addresses transactions, so mid-transaction cursors re-include their boundary transaction via PackageScanBounds::tx_range and the already-served writes are skipped during expansion — the same shape as the unfiltered ListEvents path. - The workspace temporarily patches sui-rpc (and siblings) to a local sui-rust-sdk checkout carrying the ListPackages proto; the patch is removed once that PR merges and the pinned rev is bumped.

Commit:a1c4d1f
Author:Nick
Committer:Nick

tmp

Commit:e1bd846
Author:Nick
Committer:Nick

grpc-loadtest: move build context out of docker/ (meta-svc ignores .dockerignore negations)

Commit:b78d388
Author:Nick
Committer:Nick

grpc-loadtest: add k6 build context for meta-svc

Commit:d9eb87d
Author:wlmyng
Committer:GitHub

cherry-pick Encapsulate event bitpacking (#27179) (#27226) ## Description Previously the LedgerService cursor was assumed to be opaque and hence could contain implementation details. It sort of graduated to a semi-public API that other internal services like GraphQL and gRPC's SubscriptionService need to understand. wlmyng@ pointed out that this sort of couples all implementors of the gRPC interface to one event bitpacking scheme, which could cause problems if we ever need to split the rocksdb/bigtable impls, or add a third impl someday on a completely different backend database. It also required gRPC SubscriptionService and GraphQL to understand the bitpacking scheme, which isn't ideal. This PR refactors so the cursor is just the checkpoint/txn/event offsets, and the encoding into the bit-packed u64 happens inside gRPC, closer to the storage layer. ## Test plan Unit tests all pass. ## Description Describe the changes or additions included in this PR. ## Test plan How did you test the new or updated feature? --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework: Co-authored-by: Nick Vikeras <nickvikeras@gmail.com>

Commit:9261ab6
Author:Nick Vikeras
Committer:Will Yang

Encapsulate event bitpacking (#27179) ## Description Previously the LedgerService cursor was assumed to be opaque and hence could contain implementation details. It sort of graduated to a semi-public API that other internal services like GraphQL and gRPC's SubscriptionService need to understand. wlmyng@ pointed out that this sort of couples all implementors of the gRPC interface to one event bitpacking scheme, which could cause problems if we ever need to split the rocksdb/bigtable impls, or add a third impl someday on a completely different backend database. It also required gRPC SubscriptionService and GraphQL to understand the bitpacking scheme, which isn't ideal. This PR refactors so the cursor is just the checkpoint/txn/event offsets, and the encoding into the bit-packed u64 happens inside gRPC, closer to the storage layer. ## Test plan Unit tests all pass.

Commit:bdca565
Author:Nick Vikeras
Committer:GitHub

Encapsulate event bitpacking (#27179) ## Description Previously the LedgerService cursor was assumed to be opaque and hence could contain implementation details. It sort of graduated to a semi-public API that other internal services like GraphQL and gRPC's SubscriptionService need to understand. wlmyng@ pointed out that this sort of couples all implementors of the gRPC interface to one event bitpacking scheme, which could cause problems if we ever need to split the rocksdb/bigtable impls, or add a third impl someday on a completely different backend database. It also required gRPC SubscriptionService and GraphQL to understand the bitpacking scheme, which isn't ideal. This PR refactors so the cursor is just the checkpoint/txn/event offsets, and the encoding into the bit-packed u64 happens inside gRPC, closer to the storage layer. ## Test plan Unit tests all pass.

Commit:96c233d
Author:Nick
Committer:Nick

Reserve legacy cursor proto tags Tags 1-4 carried the previous cursor schema, which is live in testnet GraphQL. Reserving them moves the new fields to 5-8 so stale tokens decode to an empty message and are rejected as missing a position, rather than depending on wire-type mismatch for clean failure.

Commit:5e0d3a4
Author:Nick
Committer:Nick

Encapsulate event bitpacking

Commit:74fa583
Author:Nick

Encapsulate event bitpacking

Commit:11ddb1d
Author:Nick

grpc-loadtest: move build context out of docker/ (meta-svc ignores .dockerignore negations)

Commit:a5c4667
Author:Nick

grpc-loadtest: add k6 build context for meta-svc

Commit:5f20508
Author:wlmyng
Committer:GitHub

[graphql] Migrate Transaction type to use gRPC cursor format (#27035)

Commit:3e0b5d4
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: record package publish checkpoint for point-in-time reads Previously the `package_versions` CF mapped `(original_id, version)` to just the storage id, with no record of when each version was published — so it could not resolve the latest version of a package as of a past checkpoint, which GraphQL needs. Add an optional publish `checkpoint` to `PackageVersionInfo`. Tip indexing records the real checkpoint; the live-set restore writes its rows via the new `store_restored` with the checkpoint left unset (a restore floor), marking versions published before the available window. The new `get_package_at_checkpoint` walks a package's versions and returns the highest whose publish checkpoint is at or before the queried one, treating restore floors as having always existed. Regenerated the proto code with `codegen.rs` and added read-query tests.

Commit:83c27de
Author:Will Yang
Committer:Will Yang

Use proto. bootstrap test so ci can catch drift, and use cargo test -p sui-rpc-cursor to regen migrate cursor-related tests to rust redundant cursor deref, reduce pagination size

Commit:84aad6d
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: record package publish checkpoint for point-in-time reads Previously the `package_versions` CF mapped `(original_id, version)` to just the storage id, with no record of when each version was published — so it could not resolve the latest version of a package as of a past checkpoint, which GraphQL needs. Add an optional publish `checkpoint` to `PackageVersionInfo`. Tip indexing records the real checkpoint; the live-set restore writes its rows via the new `store_restored` with the checkpoint left unset (a restore floor), marking versions published before the available window. The new `get_package_at_checkpoint` walks a package's versions and returns the highest whose publish checkpoint is at or before the queried one, treating restore floors as having always existed. Regenerated the proto code with `codegen.rs` and added read-query tests.

Commit:c143feb
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: resolve checkpoint-pinned reads below the restore anchor Change the `object_version_by_checkpoint` value from a bare version varint to a `Protobuf<ObjectVersionInfo>` carrying the version plus an `optional bool from_restore`. The flag is set only on rows written by the live-set restore at the anchor checkpoint (`store_restored`); tip indexing and backfill leave it unset (`store`), so the common case pays no extra bytes. Extend `get_object_version_at_checkpoint` with a restore fallback. When no entry is at or before the queried checkpoint, the object's earliest row is the only candidate: a restore-floor row (`from_restore` set) means the object predated the available window and was live then, so return it; any other earliest row is a later change or creation, so the object did not exist as of the queried checkpoint. This is what lets a static object's restore floor and an object created in the anchor checkpoint -- which share a single row at the anchor -- be told apart. The function is correct over `[T, ...]` today, where `T` is the restore anchor. Serving `[L, T)` in embedded mode additionally needs the index backfilled over `[L, T]` with synthetic floor rows for objects that change in the window, and its `reader_lo` advertised at `L` -- left to a follow-up.

Commit:9fc9de6
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: record package publish checkpoint for point-in-time reads Previously the `package_versions` CF mapped `(original_id, version)` to just the storage id, with no record of when each version was published — so it could not resolve the latest version of a package as of a past checkpoint, which GraphQL needs. Add an optional publish `checkpoint` to `PackageVersionInfo`. Tip indexing records the real checkpoint; the live-set restore writes its rows via the new `store_restored` with the checkpoint left unset (a restore floor), marking versions published before the available window. The new `get_package_at_checkpoint` walks a package's versions and returns the highest whose publish checkpoint is at or before the queried one, treating restore floors as having always existed. Regenerated the proto code with `codegen.rs` and added read-query tests.

Commit:df3c818
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: resolve checkpoint-pinned reads below the restore anchor Change the `object_version_by_checkpoint` value from a bare version varint to a `Protobuf<ObjectVersionInfo>` carrying the version plus an `optional bool from_restore`. The flag is set only on rows written by the live-set restore at the anchor checkpoint (`store_restored`); tip indexing and backfill leave it unset (`store`), so the common case pays no extra bytes. Extend `get_object_version_at_checkpoint` with a restore fallback. When no entry is at or before the queried checkpoint, the object's earliest row is the only candidate: a restore-floor row (`from_restore` set) means the object predated the available window and was live then, so return it; any other earliest row is a later change or creation, so the object did not exist as of the queried checkpoint. This is what lets a static object's restore floor and an object created in the anchor checkpoint -- which share a single row at the anchor -- be told apart. The function is correct over `[T, ...]` today, where `T` is the restore anchor. Serving `[L, T)` in embedded mode additionally needs the index backfilled over `[L, T]` with synthetic floor rows for objects that change in the window, and its `reader_lo` advertised at `L` -- left to a follow-up.

Commit:3b2b6cf
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: Index end-of-epoch system info into the epochs CF Previously the `epochs` CF recorded only `end_timestamp_ms` and `end_checkpoint` for an epoch's end, leaving it short of the data postgres keeps in `kv_epoch_ends`. This commit brings the CF to parity with the postgres epoch tables so the same epoch fields can be served later. Extend `StoredEpoch` with the end-of-epoch fields: `tx_hi` (the epoch's `network_total_transactions`), `safe_mode`, the `SystemEpochInfoEvent` gas and stake counters (`total_stake`, `storage_fund_balance`, `storage_fund_reinvestment`, `storage_charge`, `storage_rebate`, `stake_subsidy_amount`, `total_gas_fees`, `total_stake_rewards_distributed`, `leftover_storage_fund_inflow`), and the BCS-encoded `epoch_commitments`. The field-wise merge operator copies each new field when present, and the epochs pipeline now extracts them from the boundary checkpoint -- mirroring the postgres `kv_epoch_ends` handler -- recording `safe_mode = true` and leaving the counters unset when no `SystemEpochInfoEvent` is emitted. These fields are written and stored but not yet surfaced through a read path; `get_epoch` is unchanged. Regenerated the proto code with `codegen.rs`, and added merge-storage and event-extraction tests.

Commit:52f433d
Author:Brandon Williams

rpc-store: Index end-of-epoch system info into the epochs CF Previously the `epochs` CF recorded only `end_timestamp_ms` and `end_checkpoint` for an epoch's end, leaving it short of the data postgres keeps in `kv_epoch_ends`. This commit brings the CF to parity with the postgres epoch tables so the same epoch fields can be served later. Extend `StoredEpoch` with the end-of-epoch fields: `tx_hi` (the epoch's `network_total_transactions`), `safe_mode`, the `SystemEpochInfoEvent` gas and stake counters (`total_stake`, `storage_fund_balance`, `storage_fund_reinvestment`, `storage_charge`, `storage_rebate`, `stake_subsidy_amount`, `total_gas_fees`, `total_stake_rewards_distributed`, `leftover_storage_fund_inflow`), and the BCS-encoded `epoch_commitments`. The field-wise merge operator copies each new field when present, and the epochs pipeline now extracts them from the boundary checkpoint -- mirroring the postgres `kv_epoch_ends` handler -- recording `safe_mode = true` and leaving the counters unset when no `SystemEpochInfoEvent` is emitted. These fields are written and stored but not yet surfaced through a read path; `get_epoch` is unchanged. Regenerated the proto code with `codegen.rs`, and added merge-storage and event-extraction tests.

Commit:27474e2
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: resolve checkpoint-pinned reads below the restore anchor Change the `object_version_by_checkpoint` value from a bare version varint to a `Protobuf<ObjectVersionInfo>` carrying the version plus an `optional bool from_restore`. The flag is set only on rows written by the live-set restore at the anchor checkpoint (`store_restored`); tip indexing and backfill leave it unset (`store`), so the common case pays no extra bytes. Extend `get_object_version_at_checkpoint` with a restore fallback. When no entry is at or before the queried checkpoint, the object's earliest row is the only candidate: a restore-floor row (`from_restore` set) means the object predated the available window and was live then, so return it; any other earliest row is a later change or creation, so the object did not exist as of the queried checkpoint. This is what lets a static object's restore floor and an object created in the anchor checkpoint -- which share a single row at the anchor -- be told apart. The function is correct over `[T, ...]` today, where `T` is the restore anchor. Serving `[L, T)` in embedded mode additionally needs the index backfilled over `[L, T]` with synthetic floor rows for objects that change in the window, and its `reader_lo` advertised at `L` -- left to a follow-up.

Commit:6637e98
Author:Brandon Williams
Committer:GitHub

rpc-store: introduce sui-rpc-store crate (#26978) A storage backend for `sui-rpc-api`, built on `sui-consistent-store`, that hosts the column families backing every read the RPC service performs. It consolidates raw chain data previously served by the validator's perpetual / checkpoint / committee stores together with the indexes previously served by `sui-core::rpc_index` and `sui-indexer-alt-consistent-store`. The crate is organized into: - `schema` — the column-family layout. Each CF declares its on-disk key and value types plus per-CF RocksDB options. Covers raw chain data (objects, transactions, effects, events, checkpoints) and the indexes (owner, dynamic field, coin balance, package versions, epochs, ledger history, transaction / event bitmaps). - `indexer` — sequential `sui-indexer-alt-framework` pipelines that populate the schema from observed checkpoints, the `Indexer` orchestrator that wires them against a shared `Synchronizer`, plus the `pruner` and `restore` drivers for history pruning and bulk-load from a formal snapshot. - `reader` — `RpcStoreReader`, which exposes the schema through the trait stack `sui-rpc-api` consumes (`ObjectStore`, `ReadStore`, `ChildObjectResolver`, `RpcStateReader`, and `RpcIndexes`). Generic over a `Reader` so one type serves both tip reads and reads bound to a captured snapshot. - `proto` — bespoke protobuf value messages defined under `proto/sui/rpc_store/`, generated via `codegen.rs`. - `config` — `ServiceConfig` and the layered pipeline / committer / pruner / restore configuration. `start_indexer` is the standalone-binary entry point; the embedded fullnode path constructs `Indexer::from_store` directly with its own ingestion and streaming clients.

Commit:9c245c2
Author:Brandon Williams
Committer:Brandon Williams

rpc-store: introduce sui-rpc-store crate A storage backend for `sui-rpc-api`, built on `sui-consistent-store`, that hosts the column families backing every read the RPC service performs. It consolidates raw chain data previously served by the validator's perpetual / checkpoint / committee stores together with the indexes previously served by `sui-core::rpc_index` and `sui-indexer-alt-consistent-store`. The crate is organized into: - `schema` — the column-family layout. Each CF declares its on-disk key and value types plus per-CF RocksDB options. Covers raw chain data (objects, transactions, effects, events, checkpoints) and the indexes (owner, dynamic field, coin balance, package versions, epochs, ledger history, transaction / event bitmaps). - `indexer` — sequential `sui-indexer-alt-framework` pipelines that populate the schema from observed checkpoints, the `Indexer` orchestrator that wires them against a shared `Synchronizer`, plus the `pruner` and `restore` drivers for history pruning and bulk-load from a formal snapshot. - `reader` — `RpcStoreReader`, which exposes the schema through the trait stack `sui-rpc-api` consumes (`ObjectStore`, `ReadStore`, `ChildObjectResolver`, `RpcStateReader`, and `RpcIndexes`). Generic over a `Reader` so one type serves both tip reads and reads bound to a captured snapshot. - `proto` — bespoke protobuf value messages defined under `proto/sui/rpc_store/`, generated via `codegen.rs`. - `config` — `ServiceConfig` and the layered pipeline / committer / pruner / restore configuration. `start_indexer` is the standalone-binary entry point; the embedded fullnode path constructs `Indexer::from_store` directly with its own ingestion and streaming clients.

Commit:43aa508
Author:Brandon Williams
Committer:Brandon Williams

consistent-store: introduce sui-consistent-store crate A type-safe RocksDB wrapper that will back Sui's on-disk indexes, intended to replace the ad-hoc `typed-store`-style patterns in `sui-core::rpc_index` and `sui-indexer-alt-consistent-store` over time. The crate provides: - `Encode` and `Decode` traits, plus a `Protobuf<T>` wrapper that forwards them through `prost::Message`. Encoding is append-only via `BufMut` so multiple values can share one buffer. - `Db`, a cheap-to-clone handle around an opened RocksDB database carrying a bounded in-memory snapshot buffer keyed by checkpoint. - `Schema` and `SchemaAtSnapshot` for declaring column families and re-binding a struct of typed handles at a captured snapshot. The framework's bookkeeping CFs (restore state, watermark, chain id) are auto-registered and reachable via `Db::framework`. - `DbMap<K, V, R>`, the primary read/write surface — typed point gets (with zero-copy `get_raw` backed by `bytes::Bytes`), typed iteration with prefix-typed bounds, and the staging methods `Batch` uses. Parameterized by a sealed `Reader` so the same handle serves both live-tip and snapshot-bound reads, including `&Db` / `&Snapshot` zero-`Arc`-bump variants. - `Batch`, the atomic typed write batch: `put`, `delete`, `merge` across one or more CFs, committed together. - An indexer-alt-framework adapter: `Store<S>` + `Connection<'_, S>` implement `Store` and `SequentialStore` and route pipelines through a `Synchronizer` that coordinates writes from multiple sequential pipelines and takes cross-pipeline snapshots at stride boundaries. `Restore` is the sibling shape for bulk-load drivers (formal snapshot, perpetual store). - `ColumnFamilyStatsCollector`, a `prometheus::Collector` that emits per-CF RocksDB integer properties as labeled gauges, holding a weak handle so the collector does not pin the database alive. Internal Sui dependencies are limited to the indexer-alt-framework store traits and `sui-types`; the rest is `rocksdb` (with snappy, lz4, zstd, zlib, and multi-threaded CF support), `prost`, `bytes`, `parking_lot`, and `tracing`. Tests run against real `tempfile`-backed databases.

Commit:8eccec6
Author:Brandon Williams
Committer:Brandon Williams

sui-rpc-api: replace alpha proof/event services with v2alpha Previously, sui-rpc-api shipped its own alpha `ProofService` and `EventService` to back the in-tree authenticated-events light client. With the v2alpha `ProofService` (`GetCheckpointObjectProof`, with both inclusion and non-inclusion proofs) and `LedgerService.ListEvents` (filtered by `EventStreamHeadFilter`) now defined in the `sui-rpc` SDK, the in-tree stubs are obsolete. With this commit: - the alpha proof and event services are removed from `sui-rpc-api` and the v2alpha `ProofService` is mounted instead, backed by `sui_sdk_types::merkle::MerkleTree` so the SDK-side proto conversions carry the wire representation; - the rpc-index column family `events_by_stream` plus its compaction filter, `EventIndexKey`, `event_iter`, and `authenticated_event_iter` are dropped, along with `RpcConfig::authenticated_events_indexing`; authenticated event lookups now go through the existing `ledger_history_indexing` bitmap path, whose backfill already maintains the `EventStreamHead` bitmap dimension; - the in-tree prototype `sui-light-client` is ported to the new APIs, with cursor-based pagination buffering events across pages until each containing checkpoint is fully covered before folding into the MMR. The on-chain MMR is one fold per checkpoint, so partial-checkpoint folds would never match the on-chain head; - the e2e tests for authenticated events and object inclusion proofs are rewritten against the v2alpha surface, and the `corrupt_authenticated_event` fail-point moves into the new `render_event_chunk` path so the existing MMR-mismatch simtest keeps working.

Commit:3a97028
Author:stefan-mysten
Committer:GitHub

[forking] allow impersonating a sender 17/n (#26485) ## Description This PR allows to execute a transaction as an address for which we do not have the keys. The Sui CLI now has a `--forking-mode` flag to bypass the signatures check before submitting the transaction execution. Finally, it renames the crate to `sui-fork`. ## Test plan Existing CI --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [x] CLI: Sui CLI has a new `--forking-mode` flag to be used with the new `sui-fork` tool to allow impersonate a sender. - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:fbb66b7
Author:Nick
Committer:Nick

Bigtable Bitmap Indexes

Commit:0af885e
Author:stefan-mysten
Committer:GitHub

[forking 10/n] add forking service grpc for interacting with the network (#26328) ## Description This PR adds a proto description for the ForkingService that enables users to interact with the forked network through the common RPC apis. The main new APIs are: ``` sui.forking.ForkingService/GetStatus sui.forking.ForkingService/AdvanceClock sui.forking.ForkingService/AdvanceCheckpoint ``` The changes to `sui-rpc-api` are now pushed in a different PR (https://github.com/MystenLabs/sui/pull/26327), so I plan to merge that first before this one, and remove the changes here. ## Test plan Existing tests. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:45720b7
Author:Nick
Committer:Nick

Bigtable Bitmap Indexes

Commit:219dd88
Author:alex-mysten
Committer:alex-mysten

Switch ProofService and EventService from alpha to v2 proto definitions

Commit:5b9dfa1
Author:alex-mysten
Committer:Alexander Stek

Switch ProofService and EventService from alpha to v2 proto definitions

Commit:df0b148
Author:wlmyng
Committer:GitHub

[jsonrpc-alt] Migrate to read from consistent-store instead of `obj_info` (#25159) ## Description jsonrpc-alt uses `obj_info` to filter for owned objects by a type prefix, and to support queries for owned objects not of a certain type (slush requirement.) However, the latter is quite inefficient as it requires a sequential scan through the postgres db, and we have a consistent-store that can satisfy these queries, so this PR migrates jsonrpc-alt over to read from consistent-store instead of `obj_info` to unblock getting rid of the last consistent pipelines in postgres. This PR implements a basic form of `MatchNone`, which just supports "owned objects not of this type". We plan to support type inclusion/exclusion filters eventually. From consistent-store up: - `skip_prefix` in the forward and reverse iter to skip past a provided prefix - in pagination.rs, `paginate_from_front` and its equivalent additionally skips past entries starting with `exclude_prefix` - in consistent-store's `list_owned_objects.rs`, we assume a base prefix of `OwnerKind`, and a single exclusion filter on object type - `ListOwnedObjectsRequest` has new field `optional string exclude_object_type = 3;` - consistent service returns new error if both `object_type` and `exclude_object_type` are provided - consistent reader `list_owned_objects` method has additional `exclude_object_type` ## Test plan Existing tests in `jsonrpc_compound_object_filter.rs`, unit tests in `iter.rs` and `pagination.rs`. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:68db897
Author:Andrew Schran
Committer:GitHub

Add per-commit binary logging for congestion control (#25398) ## Description - Includes per-object and per-tx cost details. - Saves logged proto files to disk. - Default size limit 10x 100MB files. ## Test plan Unit tests --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:1e1c68e
Author:Alex Stek | Mysten Labs
Committer:GitHub

[light-client] authenticated events client (#24226) ## Description Client for authenticated events, which allows joining an event stream at the present moment and resuming an event stream from a previously validated checkpoint (in case of interruption). A few reviewer considerations: 1. At present we do not have a mechanism for replaying archival (pruned) events. This client solely allows joining (and rejoining, in case of temporary interruption) a stream of incoming events. To date, retrieving archival events still requires downloading and processing checkpoints. 2. Trust ratcheting requires that we fetch the last checkpoint of the epoch to verify the next epoch's committee, from genesis to the current epoch. This presumes that the checkpoint data is unpruned on the full node. 3. Resuming a stream requires that the last checkpoint is one where events were emitted and the EventStreamHead was updated. The use case is the scenario where a client goes down and needs to resume from the last verified checkpoint where we received events. It is not currently possible to start the stream from an arbitrary unpruned checkpoint. 4. This light client is tightly coupled to sui_types and other dependencies within the sui repo. We should consider the right long-term home for this implementation (sui-rust-sdk?) ## Test plan New e2e tests added. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK:

Commit:7267c60
Author:wlmyng
Committer:GitHub

Cherry pick #24963 exposing address balances from consistent service up to graphql (#25025) ## Description New `Balance.address_balance` and `Balance.coin_balance` fields to reflect the two possible sources of balance for an address. `Balance.total_balance` will return the sum of address and coin balances. ## Test plan How did you test the new or updated feature? --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [x] GraphQL: `Balance.totalBalance` now returns the sum of balances from owned coins and from the accumulator object. The individual coin or address balances can be retrieved through `Balance.coinBalance` and `Balance.addressBalance` respectively. For the previous behavior, select the `Balance.coinBalance` field for coin balances only. - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:ff5565a
Author:wlmyng
Committer:GitHub

Expose address balances from consistent service up through graphql (#24963)

Commit:7780807
Author:wlmyng
Committer:GitHub

[consistent-store] Index address balances (#24713)

Commit:1baebbb
Author:Alex Stek | Mysten Labs
Committer:GitHub

[light-client] index accumulator version on events (#24304) ## Description Index accumulator version on events. When building the MMR for authenticated event verification we need to group events by commit rather than checkpoint as there may be multiple commits / settlements per checkpoint. ## Test plan Added a simtest to create multiple commits per checkpoint. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: - [ ] Indexing Framework:

Commit:13b6a21
Author:stefan-mysten
Committer:GitHub

[pkg-alt] merge main into sui-pkg-alt (#24395)

Commit:0245ae7
Author:Alex Stek | Mysten Labs
Committer:GitHub

get_obj_inclusion_proof API and e2e light-client test (#23960) ## Description Add a new API to fetch an object version at a given checkpoint sequence number along with an inclusion proof the object was written during that checkpoint sequence number. ## Test plan Light client e2e tested + get_object_inclusion_proof happy / unhappy cases --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK:

Commit:9592048
Author:Brandon Williams
Committer:GitHub

chore: update tonic to 0.14 (#24032) This patch updates tonic to 0.14 as well as adds support for building the bigtable client from proto files directly (instead of just having a the generated protos checked in).

Commit:ff78ac0
Author:Alexander Stek
Committer:Alexander Stek

get_obj_inclusion_proof API and e2e light-client test

Commit:55c49bd
Author:Alexander Stek
Committer:Alexander Stek

get_obj_inclusion_proof API and e2e light-client test

Commit:0f99f92
Author:Alexander Stek
Committer:Alexander Stek

get_obj_inclusion_proof API and e2e light-client test

Commit:44fce9a
Author:Alex Stek | Mysten Labs
Committer:GitHub

Authenticated Event Streams - ListAuthenticatedEvents API & index (#23560) ## Description Created a new full node gRPC API to query authenticated events given a stream_id, a checkpoint seq, and limit. Todo: Resolve API model for an event stream inclusion proof & integrate prover, add more e2e tests including proofs & multiple events per txn ## Test plan How did you test the new or updated feature? --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK:

Commit:b5fe4ed
Author:alex.stek
Committer:alex.stek

Refer to stable Event proto type; index events within txn indexing

Commit:34de0e5
Author:Alexander Stek
Committer:alex.stek

paginate based on event counts. remove proofs

Commit:0a6f5f8
Author:Alexander Stek
Committer:alex.stek

authenticated event streams RPC + Index

Commit:ffa6de4
Author:Alexander Stek
Committer:Alexander Stek

Always query on a fixed range, allow optionally specifying end checkpoint

Commit:b073de0
Author:alex.stek
Committer:Alexander Stek

Refer to stable Event proto type; index events within txn indexing

Commit:bc16ba0
Author:Alexander Stek
Committer:Alexander Stek

paginate based on event counts. remove proofs

Commit:fc9ed9a
Author:Alexander Stek
Committer:Alexander Stek

authenticated event streams RPC + Index

Commit:c45799a
Author:Alexander Stek
Committer:Alexander Stek

authenticated events light client end to end demo

Commit:589140a
Author:Alexander Stek
Committer:Alexander Stek

authenticated events light client end to end demo

Commit:018fb97
Author:Alexander Stek
Committer:alex.stek

paginate based on event counts. remove proofs

Commit:5279fe7
Author:Alexander Stek
Committer:alex.stek

authenticated event streams RPC + Index

Commit:17657a5
Author:alex.stek
Committer:alex.stek

Refer to stable Event proto type; index events within txn indexing

Commit:45edcc6
Author:alex.stek
Committer:alex.stek

Refer to stable Event proto type; index events within txn indexing

Commit:e06b61a
Author:Alexander Stek
Committer:alex.stek

paginate based on event counts. remove proofs

Commit:e9dea28
Author:Alexander Stek
Committer:alex.stek

authenticated event streams RPC + Index

Commit:dba3ca9
Author:alex.stek
Committer:alex.stek

Refer to stable Event proto type; index events within txn indexing

Commit:a33e51b
Author:Alexander Stek
Committer:alex.stek

paginate based on event counts. remove proofs

Commit:889a1cf
Author:Alexander Stek
Committer:alex.stek

authenticated event streams RPC + Index

Commit:931e812
Author:Alexander Stek
Committer:Alexander Stek

paginate based on event counts. remove proofs

Commit:ecfc78d
Author:Alexander Stek
Committer:Alexander Stek

authenticated event streams RPC + Index

Commit:c6a9533
Author:Alexander Stek
Committer:Alexander Stek

paginate based on event counts. remove proofs

Commit:86a00f4
Author:Alexander Stek
Committer:Alexander Stek

authenticated event streams RPC + Index

Commit:6c01876
Author:Alexander Stek
Committer:Alexander Stek

authenticated event streams RPC + Index

Commit:12839c0
Author:Alexander Stek

switch from using transaction Digest to idx

Commit:de847e0
Author:Alexander Stek
Committer:Alexander Stek

rename api to ListAuthenticatedEvents to conform to RPC

Commit:dfcb1b2
Author:Alexander Stek

rename Event

Commit:617d9df
Author:Alexander Stek
Committer:Alexander Stek

generate rpc api within monorepo

Commit:583ea9a
Author:Tim Zakian
Committer:GitHub

Merge `main` at `91a15fe19e` into `bella-ciao` branch (#23430) ## Description Merge of `main` (at `91a15fe19e`) into the `bella-ciao` branch. NB: I _have not_ merged anything outside of `external-crates` `sui-types` and `sui-move-natives` pretty much in this PR. In particular, you will still see `HEAD` markers in e.g., the adapter or possibly other places. This will be handled in a separate PR (and we can discuss more at the bella-ciao meeting later today). ## Test plan Make sure `external-crates` is green, rest will be red. --- ## Release notes Check each box that your changes affect. If none of the boxes relate to your changes, release notes aren't required. For each box you select, include information after the relevant heading that describes the impact of your changes that a user might notice and any actions they must take to implement updates. - [ ] Protocol: - [ ] Nodes (Validators and Full nodes): - [ ] gRPC: - [ ] JSON-RPC: - [ ] GraphQL: - [ ] CLI: - [ ] Rust SDK: --------- Signed-off-by: longhutianjie <keplrnewton@icloud.com> Signed-off-by: liuchuan <582041246@qq.com> Co-authored-by: Ashok Menon <ashok@mystenlabs.com> Co-authored-by: Brandon Williams <brandon@mystenlabs.com> Co-authored-by: stefan-mysten <135084671+stefan-mysten@users.noreply.github.com> Co-authored-by: Mark Logan <103447440+mystenmark@users.noreply.github.com> Co-authored-by: wlmyng <127570466+wlmyng@users.noreply.github.com> Co-authored-by: Andrew Schran <aschran@mystenlabs.com> Co-authored-by: Todd Nowacki <tmn@mystenlabs.com> Co-authored-by: Joey Pham <joey.pham@mystenlabs.com> Co-authored-by: Xun Li <lxfind@gmail.com> Co-authored-by: Bridgerz <bridgerzoske@gmail.com> Co-authored-by: Adam Welc <adam@mystenlabs.com> Co-authored-by: Debirth <44721066+zcy1024@users.noreply.github.com> Co-authored-by: Mark Logan <mark@mystenlabs.com> Co-authored-by: Dario Russi <113150618+dariorussi@users.noreply.github.com> Co-authored-by: Alex Stek | Mysten Labs <alex.stek@mystenlabs.com> Co-authored-by: Damir Shamanaev <damirka.ru@gmail.com> Co-authored-by: Michael D. George <mike.george@mystenlabs.com> Co-authored-by: John Martin <johnjosephmartin@icloud.com> Co-authored-by: mwtian <81660174+mwtian@users.noreply.github.com> Co-authored-by: lazymio <ziqiaokong@gmail.com> Co-authored-by: Cam Swords <cameronswords@gmail.com> Co-authored-by: Deepak Maram <deepak@mystenlabs.com> Co-authored-by: phoenix <51927076+phoenix-o@users.noreply.github.com> Co-authored-by: Evan Wall <evan.wall@mystenlabs.com> Co-authored-by: akmysti <222086987+akmysti@users.noreply.github.com> Co-authored-by: Eugene Boguslavsky <eugene@mystenlabs.com> Co-authored-by: ronny-mysten <118224482+ronny-mysten@users.noreply.github.com> Co-authored-by: Rifvck Zieger <rifvckzieger@gmail.com> Co-authored-by: Jessie Mongeon <133128541+jessiemongeon1@users.noreply.github.com> Co-authored-by: Arun Koshy <97870774+arun-koshy@users.noreply.github.com> Co-authored-by: longhutianjie <keplrnewton@icloud.com> Co-authored-by: Jort <jordan.jennings@mystenlabs.com> Co-authored-by: Emma Zhong <emma@mystenlabs.com> Co-authored-by: Will Bradley <will.bradley@mystenlabs.com> Co-authored-by: Ronny Roland <ronmilton@mystenlabs.com> Co-authored-by: hoffman <bettthoffman@gmail.com> Co-authored-by: Nick Vikeras <nickvikeras@gmail.com> Co-authored-by: henryachen <henryachen93@gmail.com> Co-authored-by: cg <146006860+chris-gorham@users.noreply.github.com> Co-authored-by: Abhinav Garg <abhinavg6@gmail.com> Co-authored-by: Abhinav Garg <abhi.garg@mystenlabs.com> Co-authored-by: Radovenchyk <raddole@ukr.net> Co-authored-by: Rej Ect <99460023+rejected-l@users.noreply.github.com> Co-authored-by: Andrey Chursin <andrey@mystenlabs.com> Co-authored-by: Henry Chen <henry.chen@mystenlabs.com> Co-authored-by: Bob Liu <akagi201@gmail.com> Co-authored-by: Manolis Liolios <manos@mystenlabs.com> Co-authored-by: nikos-terzo <139557324+nikos-terzo@users.noreply.github.com> Co-authored-by: Sang Kim <nonvis@gmail.com> Co-authored-by: nikos-terzo <nikos.terzo@mystenlabs.com> Co-authored-by: leo <lc582041246@gmail.com> Co-authored-by: Mingwei Tian <mingwei@mystenlabs.com> Co-authored-by: Anastasios Kichidis <akihidis@gmail.com> Co-authored-by: Jordan Gensler <jordan@mystenlabs.com> Co-authored-by: William Robertson <williamrobertson13@users.noreply.github.com> Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Will Yang <willyang@mystenlabs.com>

The documentation is generated from this commit.

Commit:87c6030
Author:Bridgerz
Committer:Bridgerz

Add Coin Registry System Object

Commit:9242d77
Author:Ashok Menon
Committer:Ashok Menon

consistency: naming and docs changes ## Description Clarify some names and docs in the consistent store. ## Test plan Existing tests: ``` $ cargo nextest run \ -p sui-indexer-alt-e2e-tests \ -p sui-indexer-alt-consistent-api \ -p sui-indexer-alt-consistent-store \ -p sui-indexer-alt-graphql ```

The documentation is generated from this commit.