Proto commits in unkeyed/unkey

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

Commit:e1f52ff
Author:James P
Committer:GitHub

Merge branch 'main' into perkinsjr/truffle-security

Commit:76c8ec8
Author:James

feat(ctrl): disable TruffleHog-reported keys instead of deleting them A deleted key disappears from the dashboard, so a workspace cannot see which key leaked. Disable it instead, in both root key stores and for issued API keys. The key stays visible, and the audit log records a key or root key update by the system actor. A key that is already disabled still alerts the workspace, without another audit log, so a retry after an unjournaled commit never drops the alert and a workspace that disabled the key itself learns it leaked. The emails now say the key is disabled and tell the workspace to delete it rather than enable it again.

Commit:0edf527
Author:Meg Stepp
Committer:GitHub

fix(review): delete a portal's domains with it instead of refusing

Commit:3ae8b5d
Author:Meg Stepp
Committer:GitHub

fix(review): lock the portal on add, split handlers, move cname_domain under [portal]

Commit:4e3081a
Author:Meg Stepp
Committer:GitHub

fix(review): run Domain Connect discovery for portal domains

Commit:c28bc41
Author:Meg Stepp
Committer:GitHub

feat(ctrl): add the portal domain service to control-api

Commit:6f5914d
Author:Meg Stepp
Committer:GitHub

fix(frontline): match gateway policies on canonical paths (#8245) * frontline: add canonical request path helper * frontline: match policies on canonical paths and let GET cover HEAD * frontline: key per-path rate limits on the canonical path * docs: explain gateway path and method matching * frontline: skip path cleanup work for already-clean paths * frontline: reject encoded path separators instead of matching both readings * frontline: reject double-encoded path separators * docs: put stricter gateway key auth policies above broad ones * frontline: reject only ambiguous paths that can bypass a policy

Commit:cc0f1c0
Author:James
Committer:James

feat(ctrl): carry key store, kind, and revoke flag on key leak notifications FindLeakedKeysByHashes now returns the owning workspace ID, which table each match came from, and the key's kind. Key ID prefixes cannot tell unkey_root_keys from legacy keys rows. The table cannot tell a legacy root key from an API key the workspace issued to its users. Every unkey_root_keys row is a root key, and a keys row is a root key only when for_workspace_id is set. KeyLeakNotification gains workspace_id, key_store, revoke, key_hash, and key_kind. New soft-delete queries change a row only while the key is not deleted and its stored hash matches key_hash. Disabled and expired keys still match, because they can be enabled again. FindUnkeyRootKeyByID reads a root key's hash and deletion state.

Commit:795dbb0
Author:James

feat(ctrl): name the key leak reporter in Slack alerts KeyLeakNotification gains a reporter enum. The worker renders unspecified reporters as GitHub, so notifications already queued in Restate keep their current wording.

Commit:a64d13c
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:5762657
Author:Flo
Committer:Flo

ctrl: include deployment regions in connection catalogs Read caller and target placement in the same transaction as connection selection. Rebuild the catalog on region renames so Krane can publish aliases only where callers run and export endpoints where targets run. Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:a9fa6af
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:2f84ef3
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:aa9c0e7
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:51d5086
Author:Meg Stepp
Committer:GitHub

fix(review): delete a portal's domains with it instead of refusing

Commit:407ca73
Author:Meg Stepp
Committer:GitHub

feat(ctrl): add the portal domain service to control-api

Commit:753013e
Author:Meg Stepp
Committer:GitHub

fix(review): lock the portal on add, split handlers, move cname_domain under [portal]

Commit:c2925af
Author:Meg Stepp
Committer:GitHub

fix(review): run Domain Connect discovery for portal domains

Commit:0cf6bd7
Author:Meg Stepp
Committer:GitHub

feat: add the portal domain verification workflow (#8004) * feat(ctrl): add the portal domain verification workflow * fix(review): stop portal verification when the row is deleted mid-run * fix(review): share verification timing between both domain workers

Commit:22801da
Author:James

Merge remote-tracking branch 'origin/main' into perkinsjr/seadragon # Conflicts: # svc/ctrl/internal/db/querier_generated.go

Commit:028cac2
Author:Meg Stepp
Committer:GitHub

feat(ctrl): add the portal domain verification workflow

Commit:40cd9c9
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:107dbc8
Author:Flo
Committer:Flo

ctrl: include deployment regions in connection catalogs Read caller and target placement in the same transaction as connection selection. Rebuild the catalog on region renames so Krane can publish aliases only where callers run and export endpoints where targets run. Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:30f571e
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:ea4af7f
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:c5eb341
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:ff97256
Author:Meg Stepp

Merge remote-tracking branch 'origin/main' into eng-3210-gateway-path-and-method-policies-can-miss-requests-the-app

Commit:cdf90c0
Author:Meg Stepp

Merge remote-tracking branch 'origin/main' into eng-3216-frontline-policy-hardening # Conflicts: # internal/sdk/.speakeasy/gen.lock # web/internal/api/src/lib/config.ts

Commit:4238226
Author:Flo
Committer:GitHub

refactor: read live deployment pointer from environments Promote, rollback, stop, auto-promotion, custom-domain routing, the DeploymentService current-deployment guard, teardown and resume now read environments.current_deployment_id and is_rolled_back. Writers still set the app's pointer too, because the apps API, the portal and the dashboard read it. Suspension records are keyed by environment. Resume still restores the app-keyed records written before this change. Deploy after the double write and its backfill. Amp-Thread-ID: https://ampcode.com/threads/T-01a184b8-ca94-761a-8036-c5e6a7ac7522

Commit:eed1c94
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:e161e29
Author:Flo
Committer:Flo

ctrl: include deployment regions in connection catalogs Read caller and target placement in the same transaction as connection selection. Rebuild the catalog on region renames so Krane can publish aliases only where callers run and export endpoints where targets run. Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:92197e5
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:3adf4fc
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:256fe10
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:c2f5f1a
Author:Flo
Committer:GitHub

deploy: retain pinned connection targets (#7648) Defer automatic stops while current rules or running snapshots pin a revision. Serialize pin checks with stopping and promotion, and retry deferred stops. A replaced preview still stops after one minute, also when an app connection follows it. Connection creation and target selection belong to separate stack layers. Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf Co-authored-by: Amp <amp@ampcode.com>

Commit:4e1a5a5
Author:Flo
Committer:Flo

deploy: retain pinned connection targets Defer automatic stops while current rules or running snapshots pin a revision. Serialize pin checks with stopping and promotion, and retry deferred stops. A replaced preview still stops after one minute, also when an app connection follows it. Connection creation and target selection belong to separate stack layers. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:4ded215
Author:Andreas Thomas
Committer:polylane-bot

fix(deploy): refuse build forks unrelated to the app's repository Co-authored-by: polylane[bot] <277585245+polylane[bot]@users.noreply.github.com>

Commit:41f8db8
Author:Meg Stepp
Committer:GitHub

fix(review): delete a portal's domains with it instead of refusing

Commit:31bfdc9
Author:Meg Stepp
Committer:GitHub

fix(review): run Domain Connect discovery for portal domains

Commit:82431aa
Author:Meg Stepp
Committer:GitHub

fix(review): lock the portal on add, split handlers, move cname_domain under [portal]

Commit:bc31cfa
Author:Meg Stepp
Committer:GitHub

feat(ctrl): add the portal domain service to control-api

Commit:25e5815
Author:Meg Stepp
Committer:GitHub

feat(ctrl): add the portal domain verification workflow

Commit:c8e3f37
Author:Andreas Thomas
Committer:GitHub

feat(ctrl): re-encrypt recoverable keys with a Restate cron (#8205) * feat(ctrl): re-encrypt recoverable keys with a Restate cron Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a * refactor(ctrl): remove batch helper from re-encryption cron * fix(ctrl): retry and instrument vault data re-encryption Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a Co-authored-by: Amp <amp@ampcode.com> --------- Co-authored-by: Amp <amp@ampcode.com>

Commit:494e82c
Author:Meg Stepp
Committer:GitHub

frontline: reject only ambiguous paths that can bypass a policy

Commit:838853a
Author:Meg Stepp
Committer:GitHub

frontline: match policies on canonical paths and let GET cover HEAD

Commit:3ac4b60
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:bee3144
Author:Flo
Committer:Flo

ctrl: include deployment regions in connection catalogs Read caller and target placement in the same transaction as connection selection. Rebuild the catalog on region renames so Krane can publish aliases only where callers run and export endpoints where targets run. Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:28f564d
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:18b172b
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:3a57e7c
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:0990f8b
Author:Flo

krane: publish connections only where the caller or target runs Every Krane on a platform published every caller's ExternalName Services, policies, and namespaces, so each cluster's object count grew with all callers on the platform. Kubernetes recommends at most 10,000 Services and 10,000 namespaces per cluster, and every node's kubelet and Cilium agent cache every ExternalName Service. Ctrl now sends the regions where each caller and target deployment is meant to run. Krane publishes a connection's ExternalName Services only in the caller's regions, and its shared target Service and policy in the caller's and the target's regions: the target's cluster must export its endpoints and admit the remote caller. Connections that run in neither are left to certified cleanup, which also removes objects published before this change. Renaming a region now rebuilds the catalog, because connections carry region names. Amp-Thread-ID: https://ampcode.com/threads/T-01a19bcc-5ecc-7769-8f7d-0632bdd9a186

Commit:08730a3
Author:Flo
Committer:GitHub

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:de413df
Author:Flo
Committer:GitHub

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:05729cb
Author:Flo
Committer:GitHub

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:21de033
Author:Flo
Committer:GitHub

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:8642e30
Author:Flo
Committer:GitHub

deploy: retain pinned connection targets Defer automatic stops while current rules or running snapshots pin a revision. Serialize pin checks with stopping and promotion, and retry deferred stops. A replaced preview still stops after one minute, also when an app connection follows it. Connection creation and target selection belong to separate stack layers. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:64caf9f
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:5305ffe
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:c09b43e
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:0025d48
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:a16b70e
Author:Flo
Committer:Flo

deploy: retain pinned connection targets Defer automatic stops while current rules or running snapshots pin a revision. Serialize pin checks with stopping and promotion, and retry deferred stops. A replaced preview still stops after one minute, also when an app connection follows it. Connection creation and target selection belong to separate stack layers. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:c9feb76
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:f582497
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:7d80417
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:447000f
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:9c38fcd
Author:Flo
Committer:Flo

deploy: retain pinned connection targets Defer automatic stops while current rules or running snapshots pin a revision. Serialize pin checks with stopping and promotion, and retry deferred stops. A replaced preview still stops after one minute, also when an app connection follows it. Connection creation and target selection belong to separate stack layers. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:d303b77
Author:James P
Committer:GitHub

Merge branch 'main' into perkinsjr/seadragon

Commit:1b3c3ab
Author:Meg Stepp
Committer:Meg Stepp

frontline: reject only ambiguous paths that can bypass a policy

Commit:166dec1
Author:Meg Stepp
Committer:Meg Stepp

frontline: match policies on canonical paths and let GET cover HEAD

Commit:1d0b471
Author:Andreas Thomas
Committer:GitHub

feat(ctrl): re-encrypt recoverable keys with a Restate cron Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a

Commit:23c38bb
Author:Meg Stepp
Committer:Meg Stepp

frontline: reject only ambiguous paths that can bypass a policy

Commit:37f3a19
Author:Andreas Thomas
Committer:Andreas Thomas

fix(vault): fail bulk re-encryption on item errors https: //ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a Co-authored-by: Amp <amp@ampcode.com>

Commit:910cbd8
Author:Andreas Thomas
Committer:Andreas Thomas

feat(ctrl): re-encrypt recoverable keys with a Restate cron Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a

Commit:7246d3f
Author:Andreas Thomas
Committer:GitHub

feat(vault): replace single-item re-encryption with a bulk RPC (#8229) Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a Co-authored-by: Amp <amp@ampcode.com>

Commit:70948e9
Author:James

fix(ctrl): confirm workspace slug instead of delete protection on GDPR erasure

Commit:1503b47
Author:James

feat(ctrl): add GdprErasureWorkflow proto

Commit:8d2f49d
Author:Meg Stepp
Committer:GitHub

frontline: match policies on canonical paths and let GET cover HEAD

Commit:bef3f53
Author:Flo
Committer:Flo

keyleak: disable leaked keys in durable batches Move revocation before notifications into an additive Restate handler so database failures retry without blocking the webhook or changing existing journals. Batch key updates and audit inserts in one transaction. Add GitHub source metadata and replace the webhook constructor arguments with a config struct. Amp-Thread-ID: https://ampcode.com/threads/T-01a17e74-f608-7035-a75c-5a07397ea5c5

Commit:b8beb75
Author:Flo
Committer:Flo

keyleak: disable leaked keys in durable batches Move revocation before notifications into an additive Restate handler so database failures retry without blocking the webhook or changing existing journals. Batch key updates and audit inserts in one transaction. Add GitHub source metadata and replace the webhook constructor arguments with a config struct. Amp-Thread-ID: https://ampcode.com/threads/T-01a17e74-f608-7035-a75c-5a07397ea5c5

Commit:05804aa
Author:Andreas Thomas
Committer:Andreas Thomas

feat(ctrl): re-encrypt recoverable keys with a Restate cron Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a

Commit:cf30b40
Author:Andreas Thomas

feat(vault): replace single-item re-encryption with a bulk RPC Amp-Thread-ID: https://ampcode.com/threads/T-01a17962-6d27-7663-b96d-4738ed63b91a Co-authored-by: Amp <amp@ampcode.com>

Commit:6ef8221
Author:Flo

feat(heimdall): add local Cilium verdict diagnostics Exercise Cilium 1.20.2's real PRE/POST loader in an isolated Minikube cluster, including allowed and policy-denied traffic. Verify live TCX/freplace targets and observation map references rather than treating staged hooks as attached. Keep the diagnostic process and per-generation counters separate from Heimdall billing. Reuse the pinned protocol bindings without adding Cilium's module graph. Pod identity remains explicitly unverified; these counters are not invoice credits. Amp-Thread-ID: https://ampcode.com/threads/T-01a17abe-432c-7517-a287-c4840cf778a0

Commit:059bfb2
Author:Flo
Committer:Flo

notifications: expose management and durable test APIs Add tenant-scoped destination and rule management, personal preferences, and a shared recipient resolver. Encrypt write-only credentials with workspace Vault keyrings and make retries reuse their creation or test identity. Expose generated public API and SDK contracts. Preserve actionable deletion conflicts without leaking transport details. Recheck delivery expiry before provider calls and report authoritative deployment cancellation state. Amp-Thread-ID: https://ampcode.com/threads/T-01a0f3d7-c439-7462-83b9-0ad4825087f7

Commit:cc63720
Author:Flo
Committer:Flo

notifications: add durable delivery and deployment reporting Add workspace routing, encrypted delivery snapshots, channel providers, and Restate delivery workers. Route deployment reports through typed helpers and serialize shared GitHub PR comments. Keep uncertain provider outcomes explicit rather than retrying non-idempotent sends blindly. Amp-Thread-ID: https://ampcode.com/threads/T-01a0f3d7-c439-7462-83b9-0ad4825087f7 Co-authored-by: Amp <amp@ampcode.com>

Commit:ad7b8d4
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:7d02d86
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:0da61c2
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:908874d
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:eb78156
Author:Flo
Committer:Flo

deploy: retain pinned connection targets Defer automatic stops while current rules or running snapshots pin a revision. Serialize pin checks with stopping and promotion, and retry deferred stops. A replaced preview still stops after one minute, also when an app connection follows it. Connection creation and target selection belong to separate stack layers. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:df74a1d
Author:Andreas Thomas
Committer:GitHub

fix(api): make OCI app updates atomic (#8122) * fix(api): make OCI app updates atomic Validate OCI references before writes and persist app settings, OCI configuration, and audit events in one transaction. Amp-Thread-ID: https://ampcode.com/threads/T-01a16c4a-cd7e-7119-84e5-3397ea00337b Co-authored-by: Amp <amp@ampcode.com> * refactor(ctrl): remove unused OCI image source RPC Remove the unused RPC, generated clients, mock, and control-plane-only SQL query. Keep OCI source updates covered through the atomic public app update handler. Amp-Thread-ID: https://ampcode.com/threads/T-01a16c4a-cd7e-7119-84e5-3397ea00337b Co-authored-by: Amp <amp@ampcode.com> --------- Co-authored-by: Amp <amp@ampcode.com>

Commit:106ca2c
Author:Andreas Thomas
Committer:GitHub

docs(frontline): clarify whole-field header matching (#8135) Document commas as literal characters and require repeated header fields for separate values. Regenerate the Go and TypeScript comments without changing runtime behavior. Amp-Thread-ID: https://ampcode.com/threads/T-01a16c68-6a5e-734c-baa5-d257f6870bf8 Co-authored-by: Amp <amp@ampcode.com>

Commit:3a6fb38
Author:Oz
Committer:GitHub

fix(ctrl): end worker invocations on a missing row (#8155) * fix(ctrl): end worker invocations on a missing row tldr: deleting a project, app, or environment that is already gone, or processing a challenge for a deleted domain, left the Restate invocation retrying forever. restate.Run retries its closure on any non-terminal error, so a raw sql.ErrNoRows never reached the handler. The lookup kept retrying and the invocation never completed. before: two deletes of one project first completed second retrying, 6+ attempts over 20s after: both deletes complete pkg/restate/restateutil.TerminalNotFound turns a missing row into a terminal 404 inside the Run closure. The deletes treat that 404 as success, because the postcondition already holds. ProcessChallenge returns status "skipped". Tests drive a real Restate and assert through the admin query that the invocation is no longer live, so they fail on the retry loop itself rather than on an error the ingress happens to surface. Checked: app, environment, project and certificate packages pass, go build ./svc/ctrl/... clean. * fix(ctrl): audit before delete, bound cert renewals, verify claim tldr: three more ways a worker invocation could end without doing its job, found by review of the previous commit. **audit before delete.** the audit entry was a durable step after the row delete. if that step exhausted its retries the invocation restarted, found no row, and finished through the new "already deleted" path with no audit written. moving the insert ahead of the delete keeps the row in place when the audit fails, so the retry writes both. **CertificateService had no retry limit.** RenewExpiringCertificates runs on the fixed "global" key, so one invocation that never finished blocked every later renewal. bound at 10 attempts with kill. **the ACME claim result was discarded.** the claim update matched only waiting rows, so a renewal, whose challenge is already verified, changed nothing and issuance carried on anyway. the query now matches waiting or verified and reports rows affected. zero rows means another run claimed it, and the handler stops instead of issuing again. Checked: db, app, environment, project, certificate and integration pass, go build clean. The claim test fails against the old query. * fix(ctrl): bound only the renewal handler, fix stale comments tldr: the retry limit from the previous commit covered ProcessChallenge too, and a kill there could strand a certificate. it now covers only RenewExpiringCertificates. **kill stranded ProcessChallenge.** a kill skips the deferred compensation, so the challenge stayed pending. the renewal cron skips pending rows and the claim now skips them too, so the cert would never renew. ``` RenewExpiringCertificates ProcessChallenge before 10 attempts, kill 10 attempts, kill after 10 attempts, kill server default, pause ``` read back from the Restate admin API for both bindings. **cleanup.** TerminalNotFound checks sql.ErrNoRows itself instead of taking a predicate every caller passed the same. comments that described a restart that cannot happen, or the claim before it accepted verified rows, now match the code. ProcessChallenge doc and proto list "skipped". * revert(ctrl): drop the ACME claim change tldr: the claim change is not part of the missing-row fix, and it made one failure worse. back to the main behavior. ProcessChallenge is keyed by domain, so Restate already runs one call per domain at a time. two runs never race for the claim, so the rows-affected gate guarded nothing. matching verified rows moved a renewal to pending. a renewal killed in that state stays pending, the cron never lists pending rows, and the cert expires silently. on main the row stays verified and the next cron tick retries. ``` renewal killed mid-run claim change: row pending -> never retried -> cert expires main: row verified -> next cron tick retries ``` the missing-domain skip in step 1 stays. it does not depend on the claim. * revert(ctrl): drop the renewal retry limit tldr: the RenewExpiringCertificates retry limit is not part of the missing-row fix, so it goes to its own PR if we want it. a missing domain blocks only its own ProcessChallenge key, never the renewal job's global key, so the limit does not touch this bug. also restores a comment period changed by the earlier commits. * fix(ctrl): drop stale capture comments in delete handlers tldr: the audit now runs before the row delete, so the row is still there when the audit reads it. "capture before the row is deleted" no longer explains anything, and the app cascade never deleted the app row.

Commit:bc22485
Author:ogzhanolguncu

revert(ctrl): drop the ACME claim change tldr: the claim change is not part of the missing-row fix, and it made one failure worse. back to the main behavior. ProcessChallenge is keyed by domain, so Restate already runs one call per domain at a time. two runs never race for the claim, so the rows-affected gate guarded nothing. matching verified rows moved a renewal to pending. a renewal killed in that state stays pending, the cron never lists pending rows, and the cert expires silently. on main the row stays verified and the next cron tick retries. ``` renewal killed mid-run claim change: row pending -> never retried -> cert expires main: row verified -> next cron tick retries ``` the missing-domain skip in step 1 stays. it does not depend on the claim.

Commit:797184e
Author:Flo
Committer:Flo

ctrl: stream private network catalogs to Krane Serve StreamPrivateNetworkState to authenticated Krane clusters from the catalog of the cluster's platform. Send connections in pages of 1,000, then a complete chunk with the topology, or only an unchanged chunk when Krane already has that version. Report Unavailable until the catalog is published or while its CDC stream is silent. Ctrl API starts the catalogs and gives them to the cluster service. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:4fd0e65
Author:Flo
Committer:Flo

ctrl, krane: pass saved deployment capabilities Keep enrollment in the deployment capabilities message. Krane derives the fixed self hostname locally instead of receiving it over the wire. Missing capabilities leave private networking disabled. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a0867e-f00a-7718-961d-5d9cf2250eaf

Commit:6dbe6be
Author:Flo
Committer:Flo

ctrl: maintain private network catalogs from CDC Keep one catalog per platform current from a VStream of the rows that can change it, and read again only the callers a change can affect. Publish only after the initial copy completes. Certify a snapshot for deletions only while its checkpoint committed in the last ten seconds, and name it after that commit so every Ctrl instance agrees on its ID. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:6a82db9
Author:Flo
Committer:Flo

ctrl, krane: carry private networking enrollment to deployment Pods Read each deployment's environment kind, app slug, and capabilities with its topology, and add them to ApplyDeployment. A deployment created with private networking whose app slug is a DNS label gets its replica host, the app slug in the private zone. Other deployments are not enrolled. Krane labels Pods with the environment kind and rejects an enrolled deployment without a production or preview kind or a replica host. With a regional resolver IP, enrolled Pods use it for DNS and get UNKEY_PRIVATE_DOMAIN. Krane does not set a resolver IP yet, so Pods keep their normal DNS. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:06be6a9
Author:Flo
Committer:Flo

deploy: retain pinned connection targets Defer automatic stops while current rules or running snapshots pin a revision. Serialize pin checks with stopping and promotion, and retry deferred stops. A replaced preview still stops after one minute, also when an app connection follows it. Connection creation and target selection belong to separate stack layers. Co-authored-by: Amp <amp@ampcode.com> Amp-Thread-ID: https://ampcode.com/threads/T-01a13087-15cb-769b-9f13-0b8d82f5c947

Commit:3e935ff
Author:ogzhanolguncu

fix(ctrl): bound only the renewal handler, fix stale comments tldr: the retry limit from the previous commit covered ProcessChallenge too, and a kill there could strand a certificate. it now covers only RenewExpiringCertificates. **kill stranded ProcessChallenge.** a kill skips the deferred compensation, so the challenge stayed pending. the renewal cron skips pending rows and the claim now skips them too, so the cert would never renew. ``` RenewExpiringCertificates ProcessChallenge before 10 attempts, kill 10 attempts, kill after 10 attempts, kill server default, pause ``` read back from the Restate admin API for both bindings. **cleanup.** TerminalNotFound checks sql.ErrNoRows itself instead of taking a predicate every caller passed the same. comments that described a restart that cannot happen, or the claim before it accepted verified rows, now match the code. ProcessChallenge doc and proto list "skipped".

Commit:120997a
Author:Andreas Thomas
Committer:GitHub

fix(ctrl): serialize custom domain verification by hostname (#8123) Amp-Thread-ID: https://ampcode.com/threads/T-01a16c4b-1e8c-76df-abee-6f3f7d7e3bef Co-authored-by: Amp <amp@ampcode.com>