These 50 commits are when the Protocol Buffers files have changed:
| Commit: | 673453d | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Josep Maria Viñolas Auquer | |
revert(haproxy-sync): undo the apiv4 resync of the haproxy maps This reverts merge cf22531ef0, which went in before its review had finished. apiv4 goes back to its health watch, and haproxy-sync loses the WatchResync and AckResync RPCs, the resync retry loop, the certificate reuse and path checks, and the longer DomainSync deadline. The RPCs were never released and have no clients. The next commit keeps the maps across a reload inside haproxy-sync.
| Commit: | 500a124 | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | Josep Maria Viñolas Auquer | |
fix(haproxy-sync): retry a resync until apiv4 reports it succeeded A resync request was fire and forget: when the full sync failed on the apiv4 side, because the database or a gRPC call failed, nothing asked again and the maps stayed empty until the next reload. Every request now carries an id and apiv4 answers it with AckResync, with the error when the sync failed. haproxy-sync sends the request again, with a new id, when the answer is an error or does not arrive in five minutes, with an exponential backoff from one second capped at one minute and no limit of attempts, logging every failed attempt. Answers to an older id are ignored. A reload, or a watcher connecting, replaces the cycle being retried instead of adding a second one, and a cycle ends when no watcher is connected, since the next one to connect gets its own request.
| Commit: | aad0463 | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | Josep Maria Viñolas Auquer | |
fix(haproxy-sync): ask apiv4 to resync the maps after every haproxy reload The bastion subdomains, the bastion individual domains and the branding domains live in HAProxy virtual maps and runtime certificates, which only exist in the running worker. apiv4 pushed them at startup and when it saw haproxy-sync come back after an outage, but a graceful reload keeps haproxy-sync up and starts a new worker with empty maps, so nothing pushed them again and every bastion web address silently landed on the portal. haproxy-sync now offers a WatchResync server stream. It sends a request as soon as a client connects, which covers a portal restart, and again after every reload: haproxy-reload signals it with SIGHUP once the reload succeeded, and it waits for the new worker to own the admin socket before asking, so the entries do not land in the worker that is shutting down. apiv4 keeps that stream open from its lifespan, reconnecting with backoff, and runs the existing full sync for each request, logging the reason. It replaces the health watch, which only fired on an outage. apiv4 stays the only source of truth and nothing polls.
| Commit: | 5ad6136 | |
|---|---|---|
| Author: | Miriam Melina Gamboa Valdez | |
| Committer: | Miriam Melina Gamboa Valdez | |
feat(apiv4): report the build commit in the admin version endpoint The version alone does not tell which commit a branch or dev image was built from. build.sh now stamps the short commit hash into the apiv4 image; the admin version endpoint, the webapp About page and the check service and monitor output report it.
| Commit: | dede1ab | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | Josep Maria Viñolas Auquer | |
style(haproxy, haproxy-sync, common, apiv4): cut comments and docstrings to the load-bearing The main-domain certificate configuration work carried long comment blocks and multi-paragraph docstrings that restated the code, narrated the change or repeated rationale the MR already covers. They are removed; what survives is one line at the thing it constrains, kept only where a reader would get it wrong from the code alone: that names[0] fixes acme.sh's on-disk state directory and deploy key, that r.literal replaces the stored block instead of merging into it, that the account thumbprint is re-pushed before each issuance because a reload resets the proc var to the parse-time environment, that set var and experimental-mode ride one command string because the mode is per-connection, and that the renewal cron must be armed unconditionally so the branding certificates renew whether or not the main domain uses ACME. No behaviour changes: every touched Python and Go file parses to an identical tree once comments and docstrings are stripped, and the shell, proto and config diffs contain comment lines only.
| Commit: | 6dffa6d | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | Josep Maria Viñolas Auquer | |
feat(haproxy-sync): add the IssueMainDomain rpc for the deployment certificate The certificate the whole deployment is served with had no way in. Every domain haproxy-sync knows about arrives through DomainSync, which adds it to the domain map and gives it a crt-list entry -- and the main domain is neither of those things. It is routed by the base ACL, and its certificate IS the crt-list default rather than an entry in it. Pushing it through DomainSync would put the same name on two paths and a second certificate ahead of the default. So it gets its own call, whose whole job is to issue and deploy to that default PEM and touch nothing else. The tests declare no map or crt-list expectations at all, so reaching for either fails them. Two things about the shape are worth stating, because both are frozen the moment this merges and visible on the wire. names is repeated from the start. Nothing here issues more than one name, but the field number and cardinality are what cost a migration to change later, and this message is new so there is no compatibility to preserve. It is ordered, and names[0] is the primary: acme.sh indexes a certificate by its first name, which decides the on-disk state directory and the deploy key, so reordering describes a different certificate. Field numbers 4 to 7 are left free for the acme-dns credentials that follow, with a comment saying so and deliberately WITHOUT a reserved declaration. reserved means a number may never be assigned again -- it is for fields that were deleted -- so reserving these would forbid exactly the use they are being kept for. The call refuses, loudly, when an operator-supplied certificate is present in the certs directory. The crt-list is rebuilt on every boot and prefers that file, falling back to the ACME one only when it is absent, so while it exists everything issued here is written, deployed, logged as successful, and never served. An operator watching a certificate they just configured have no effect, with every line saying it worked, is the worst outcome available; an error naming the file that shadows it is the least bad one. The account travels with the request rather than coming from the container's environment, because it now lives in the database and can change without restarting anything. It reaches the issuance script through the positional arguments added for it earlier, one argument per name.
| Commit: | 13a0e44 | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | Josep Maria Viñolas Auquer | |
style(common, haproxy, haproxy-sync, apiv4, webapp): cut comments to the load-bearing The acme-dns (DNS-01) work carried long comment blocks and multi-paragraph docstrings that restated the code or repeated rationale across files: the argument that one acme-dns account holds a bounded number of TXT records so domains cannot share one appeared in five places, and that a partial credential set is a failure rather than a degraded HTTP-01 fallback in eight. Each is now stated once at the place it governs and referenced from the rest; the canonical copy of the shared-account argument lives on the proto acme_dns fields. What survives is one line at the thing it constrains: that the acme-dns password travels on stdin and never argv or the child environment because both are readable through /proc, that the issuance output is not captured into an error because acme.sh logs the password in it, that the challenge fingerprint omits the password so rotating it does not spend CA rate limit re-issuing an identical certificate, and that a DNS-01 path pushes no account thumbprint because the challenge never reaches HAProxy. Only the increment this branch adds over the certificates-config branch is touched; the comments that branch introduced are left to it. There is no behaviour change here, since every touched Python and Go file parses to an identical tree once comments and docstrings are stripped, and the shell, proto and JavaScript diffs contain comment lines only.
| Commit: | 1a7a634 | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | Josep Maria Viñolas Auquer | |
feat(proto): carry acme-dns credentials to haproxy-sync Both certificate scopes need this, not one. The main domain is what the customer asking for it is served on, but per-category branding domains are what most installations actually have several of, and leaving DomainSyncDomain out would make DNS-01 available for half the cases. The numbers were reserved by intent rather than by declaration. The change that added IssueMainDomain left 4 to 7 free with a comment saying they were for exactly this, deliberately not marking them reserved, which would have forbidden ever assigning them. 3 to 6 on DomainSyncDomain had simply never been used. Nothing already on the wire moves. Four fields rather than one blob, and all four rather than a flag plus some of them, because three of four cannot issue anything. Treating a partial set as "use DNS-01" turns a half-filled form into a failure deep inside acme.sh whose message is about DNS resolution, which sends the operator to look at their zone instead of at what they just saved. The field numbers get a test of their own. Renaming a protobuf field is free and renumbering one is not: apiv4 and haproxy-sync are separate containers that are not upgraded together, so a renumbering does not fail the call, it succeeds with the values shifted -- an issuance that authenticates with the subdomain as its password. buf breaking catches that, but only against main on the remote, so it does not protect these numbers until they are already there. The test covers that window and is worth keeping afterwards for the same reason the comment was worth writing.
| Commit: | 370f126 | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
style(haproxy, haproxy-sync, common, apiv4): cut comments and docstrings to the load-bearing The main-domain certificate configuration work carried long comment blocks and multi-paragraph docstrings that restated the code, narrated the change or repeated rationale the MR already covers. They are removed; what survives is one line at the thing it constrains, kept only where a reader would get it wrong from the code alone: that names[0] fixes acme.sh's on-disk state directory and deploy key, that r.literal replaces the stored block instead of merging into it, that the account thumbprint is re-pushed before each issuance because a reload resets the proc var to the parse-time environment, that set var and experimental-mode ride one command string because the mode is per-connection, and that the renewal cron must be armed unconditionally so the branding certificates renew whether or not the main domain uses ACME. No behaviour changes: every touched Python and Go file parses to an identical tree once comments and docstrings are stripped, and the shell, proto and config diffs contain comment lines only.
| Commit: | 80bbb12 | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | IsardVDI Admin | |
feat(haproxy-sync): add the IssueMainDomain rpc for the deployment certificate The certificate the whole deployment is served with had no way in. Every domain haproxy-sync knows about arrives through DomainSync, which adds it to the domain map and gives it a crt-list entry -- and the main domain is neither of those things. It is routed by the base ACL, and its certificate IS the crt-list default rather than an entry in it. Pushing it through DomainSync would put the same name on two paths and a second certificate ahead of the default. So it gets its own call, whose whole job is to issue and deploy to that default PEM and touch nothing else. The tests declare no map or crt-list expectations at all, so reaching for either fails them. Two things about the shape are worth stating, because both are frozen the moment this merges and visible on the wire. names is repeated from the start. Nothing here issues more than one name, but the field number and cardinality are what cost a migration to change later, and this message is new so there is no compatibility to preserve. It is ordered, and names[0] is the primary: acme.sh indexes a certificate by its first name, which decides the on-disk state directory and the deploy key, so reordering describes a different certificate. Field numbers 4 to 7 are left free for the acme-dns credentials that follow, with a comment saying so and deliberately WITHOUT a reserved declaration. reserved means a number may never be assigned again -- it is for fields that were deleted -- so reserving these would forbid exactly the use they are being kept for. The call refuses, loudly, when an operator-supplied certificate is present in the certs directory. The crt-list is rebuilt on every boot and prefers that file, falling back to the ACME one only when it is absent, so while it exists everything issued here is written, deployed, logged as successful, and never served. An operator watching a certificate they just configured have no effect, with every line saying it worked, is the worst outcome available; an error naming the file that shadows it is the least bad one. The account travels with the request rather than coming from the container's environment, because it now lives in the database and can change without restarting anything. It reaches the issuance script through the positional arguments added for it earlier, one argument per name.
| Commit: | 573c8bf | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Josep Maria Viñolas Auquer | |
chore(pkg): misc pkg/cfg, pkg/grpc, pkg/redis, pkg/proto, go.mod/go.sum + mockery config Collapsed from 10 original commits on the apiv4-integration branch. Rolls up the non-generated pkg/ updates: new Redis DB numbers in pkg/redis, pkg/cfg tweaks, pkg/grpc server helpers, haproxy_sync v1 proto definitions, Go dependency bumps (go.mod/go.sum), mockery v2 config, and buf.yaml / buf.gen.yaml codegen plugin wiring.
| Commit: | 02b3238 | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Josep Maria Viñolas Auquer | |
chore(pkg): misc pkg/cfg, pkg/grpc, pkg/redis, pkg/proto, go.mod/go.sum + mockery config Collapsed from 10 original commits on the apiv4-integration branch. Rolls up the non-generated pkg/ updates: new Redis DB numbers in pkg/redis, pkg/cfg tweaks, pkg/grpc server helpers, haproxy_sync v1 proto definitions, Go dependency bumps (go.mod/go.sum), mockery v2 config, and buf.yaml / buf.gen.yaml codegen plugin wiring.
| Commit: | 5d6a1ae | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Miriam Melina Gamboa Valdez | |
feat(haproxy-sync): support provided certificates in DomainSync Allow domains to carry optional PEM certificate data instead of always issuing via ACME. Make ACME RemoveCert tolerant so removal works uniformly for both certificate sources.
| Commit: | 3818477 | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Miriam Melina Gamboa Valdez | |
feat(haproxy-sync): report per-domain failures instead of aborting sync Extract addDomain, continue on error collecting failures into FailedDomains. Add DomainSyncError to proto and gRPC transport mapping.
| Commit: | 7e9542b | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Miriam Melina Gamboa Valdez | |
feat(haproxy,haproxy-sync): support multiple domains
| Commit: | b18cddd | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
refactor(haproxy-sync): rename from haproxy-bastion-sync to haproxy-sync
| Commit: | b203cf3 | |
|---|---|---|
| Author: | darta | |
| Committer: | Pau Abril | |
fix(haproxy): added haproxy grpc proto
The documentation is generated from this commit.
| Commit: | edddee0 | |
|---|---|---|
| Author: | darta | |
| Committer: | Pau Abril | |
feat(haproxy): implement haproxy-bastion-sync gRPC microservice Implement complete gRPC microservice for dynamic HAProxy map management, replacing inotify-based reloads with Runtime API for zero-downtime updates. Implementation includes: - Protocol definition with 6 RPC methods for dynamic map management - Configuration layer using viper for environment variables - HAProxy Runtime API socket communication layer * Unix socket with automatic reconnection * Commands: show map, add map, del map, clear map * Thread-safe operations with mutex - Map storage and sync logic * In-memory map file management * File parsing with auto-creation * Diff calculation between desired and current state * Incremental add/delete operations - gRPC service implementation * AddSubdomain, DeleteSubdomain (subdomains.map) * AddIndividualDomain, DeleteIndividualDomain (individual.map) * SyncMaps for full synchronization * GetCurrentMaps for state retrieval - Main entry point with lifecycle management * HAProxy socket connection with retry * Initial sync on startup * Graceful shutdown with signal handling Technical details: - Uses HAProxy stats socket for Runtime API - Thread-safe operations for concurrent gRPC requests - Proper error handling and comprehensive logging - Context management for graceful shutdown
| Commit: | 451ca6b | |
|---|---|---|
| Author: | Pau Abril | |
fix(api, storage): refactor storage convert to not use protobuf Refactor the storage convert to use kwargs instead of a protobuf message.
| Commit: | 15fbfd7 | |
|---|---|---|
| Author: | Pau Abril | |
Merge branch 'logout-user-when-deleting' into 'devel' feat(api, sessions): revoke user session See merge request isard/isardvdi!3185 (cherry picked from commit c92c1c0e0965d2e29e58a2b666c93616a72ebaa8) 523fdedb feat(api, sessions): revoke user session fdf8c9fe feat(api): revoke session when deleting a user 55af12ae test(sessions): add GetUserSessions tests Co-authored-by: Josep Maria Viñolas Auquer <josepmaria@isardvdi.com>
| Commit: | ce35cd6 | |
|---|---|---|
| Author: | Miriam Melina Gamboa Valdez | |
| Committer: | Miriam Melina Gamboa Valdez | |
feat(sessions): check request ip adress When creating, getting or renewing a session the request ip address is checked. If it has changed from the original session request an error will be returned.
| Commit: | 16523fc | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(sessions): add unique user
| Commit: | 29253a3 | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(sessions): initial sessions service This service will be responsible for handling the sessions in IsardVDI. It will handle the session renewal, expiration times and revocations
| Commit: | 90b640d | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(operations): add GPU protobuf definitions and multiple hypervisor operations
| Commit: | e03f7cc | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(check): add the check service
| Commit: | a3b4a98 | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(proto): add the storage jobs definition
| Commit: | 842a1cc | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(proto): add the python, javascript and docs plugins
| Commit: | 40ae6dc | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(operations): update the operations grpc interface
| Commit: | fe25f00 | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Néfix Estrada | |
feat(operations): define the operations GRPC interface
| Commit: | 45146b3 | |
|---|---|---|
| Author: | Néfix Estrada | |
Add Disks support Now you can create desktops with disks and start those Signed-off-by: Néfix Estrada <nefixestrada@gmail.com>
| Commit: | 7acd67d | |
|---|---|---|
| Author: | Néfix Estrada | |
Fixed desktopCreate and desktopStart methods Signed-off-by: Néfix Estrada <nefixestrada@gmail.com>
| Commit: | d39f298 | |
|---|---|---|
| Author: | Néfix | |
Return UserID, name and surname Signed-off-by: Néfix <nefixestrada@gmail.com>
| Commit: | 66d237f | |
|---|---|---|
| Author: | Néfix | |
| Committer: | Néfix | |
Fixed desktopbuilder and the GrapqhQL Signed-off-by: Néfix <nefixestrada@gmail.com>
| Commit: | 8e3453b | |
|---|---|---|
| Author: | Néfix | |
| Committer: | Néfix | |
Hasura Auth - Added Hasura Authentication Webhook - Continued with the GrapqhQL endpoint definition - Moved all the proto files in the common directory - Renamed the common directory to pkg, following Go's best practicies - Improved Dockerfiles Signed-off-by: Néfix <nefixestrada@gmail.com>
| Commit: | 3d3631e | |
|---|---|---|
| Author: | Néfix Estrada | |
| Committer: | Simó Albert i Beltran | |
Initial v3
| Commit: | 8245647 | |
|---|---|---|
| Author: | Néfix Estrada | |
Created the orchestrator and the random provider for it
| Commit: | 5f23e24 | |
|---|---|---|
| Author: | Néfix Estrada | |
v3 WIP (I lost all my local work :S)
| Commit: | 9c5dd68 | |
|---|---|---|
| Author: | Néfix Estrada | |
Finished the microservices. Only building hyper remains
| Commit: | 8807914 | |
|---|---|---|
| Author: | darta | |
added download function and testing large file stream with grpc
| Commit: | 6071cf5 | |
|---|---|---|
| Author: | darta | |
Merge branch 'feature/hyperfunctions' into v2
| Commit: | ae67ef7 | |
|---|---|---|
| Author: | darta | |
viewers progress
| Commit: | 209e6c3 | |
|---|---|---|
| Author: | darta | |
some errors to be fixed
| Commit: | 14b2704 | |
|---|---|---|
| Author: | darta | |
cleaned live share
| Commit: | 9e5e60a | |
|---|---|---|
| Author: | darta | |
Derivate with exec and some previous checks done
| Commit: | 0a645ab | |
|---|---|---|
| Author: | Josep Maria Viñolas Auquer | |
| Committer: | GitHub | |
Merge pull request #9 from jvinolas/feature/disk-operations Feature/disk operations
| Commit: | 5be2d2a | |
|---|---|---|
| Author: | darta | |
creates qcow2
| Commit: | a422852 | |
|---|---|---|
| Author: | darta | |
returns three xml
| Commit: | a54c82d | |
|---|---|---|
| Author: | darta | |
| Committer: | darta | |
tests
| Commit: | 867495a | |
|---|---|---|
| Author: | Néfix Estrada | |
Initial commit for the hyper microservice
| Commit: | 6598535 | |
|---|---|---|
| Author: | Néfix Estrada | |
Initial engine commit