Skip to content

v0.9.0

Choose a tag to compare

@github-actions github-actions released this 04 Nov 01:23
· 130 commits to main since this release
2ce7d6c

What's Changed

  • docs: the repository map says a scheduler pass is two transactions by @rburns629 in #80
  • chore: pin vertra 0.8.0 in the catalog template by @github-actions[bot] in #81
  • fix(console): a bot belongs to the one team that made it by @rburns629 in #82
  • fix(console): a catalog release registers products only in the teams it names by @rburns629 in #83
  • fix(console): products are registered by the catalog or an admin, not a team's maintainers by @rburns629 in #84
  • fix(gateway): a request's path can't stop traffic from being recorded by @rburns629 in #86
  • fix(kernel): steps held by a cap or key don't keep others from being claimed by @rburns629 in #87
  • fix(console): a chat model's key and the console's cloud identity go only where they were given for by @rburns629 in #88
  • fix(helm): the console may run only its image scans and catalog verifications as Jobs by @rburns629 in #89
  • fix(helm): the kernel's processes sign in to the bundled Postgres as a role of their own by @rburns629 in #90
  • fix(cli): the AWS catalog publish role trusts a job of its Environment only on main by @rburns629 in #91
  • fix(console): numeric settings are checked when it starts by @rburns629 in #93
  • fix(console): audit retention keeps what isn't copied yet, and isn't mistaken for tampering by @rburns629 in #94
  • fix(console): single sign-on linking an existing account makes it the only way in by @rburns629 in #95
  • fix(console): purging a team deletes its single sign-on mappings and schedules by @rburns629 in #96
  • fix(gateway): callers are checked as at the console: admin-set single sign-on, ordering bots, purged teams by @rburns629 in #97
  • fix(kernel): a cancelled execution's tasks can't be retried, and a failed one's waiting tasks aren't in the queue by @rburns629 in #98
  • fix(helm): the policies' kept lists are written as JSON strings, so an apostrophe can't break an upgrade by @rburns629 in #99
  • fix(console): single sign-on's session is taken only by the tab that started the sign-in by @rburns629 in #92
  • fix(console): a bot's run isn't approved by whoever made the bot something to act with by @rburns629 in #85
  • fix(console): a run's approvers are worked out before the kernel has it by @rburns629 in #100
  • fix(console): no bot's keyless identity may overlap another bot's by @rburns629 in #101
  • fix: an API may only use an in-cluster backend in its own team's namespace by @rburns629 in #102
  • fix(helm): a Group purger's members are held to the purge guard too by @rburns629 in #103
  • fix: the gateway's token fence bounds audience and lifetime, and -api is reserved for APIs by @rburns629 in #104
  • fix(helm): a step's container is held to what the executor builds, and hardened per container by @rburns629 in #105
  • fix(helm): the sleeper may only suspend, resume and reschedule its CronJobs (H-7) by @rburns629 in #106
  • docs: what a platform namespace shared with a team gives up (H-9) by @rburns629 in #107
  • fix(helm): the bundled databases' passwords are kept with their volumes (H-10) by @rburns629 in #108
  • fix(gateway): a path naming no API served here is counted under api="-" (G-7) by @rburns629 in #109
  • fix(gateway): a rotated-out session key stops working within a minute, on both listeners (G-8) by @rburns629 in #110
  • fix(gateway): what an API's OpenAPI document doesn't declare isn't forwarded (G-5) by @rburns629 in #111
  • fix(console): a schedule due in the hour the clocks repeat runs once (D-4) by @rburns629 in #112
  • fix(console): one console at a time makes each delivery pass (C-6) by @rburns629 in #113
  • fix(console): the resource recorder pages past runs that haven't finished (D-6) by @rburns629 in #114
  • fix(cli): destroy keeps only what's in the state (L-3) by @rburns629 in #116
  • fix(ci): the proto check covers the kernel's generated code too (X-3) by @rburns629 in #117
  • fix(cli): the aws check asks the region chosen, not only its list (M-2) by @rburns629 in #121
  • fix(catalog-template): http-check's token goes only to the URL it was given (T-1) by @rburns629 in #129
  • fix(cli): no AWS install keeps the account's GitHub OIDC provider in its state, so its destroy can't delete it from under others (M-3) by @rburns629 in #118
  • fix(cli): a location group the gcp check can't read isn't a refusal (M-1) by @rburns629 in #119
  • fix(cli): the cost screen prices the platform nodes its requests need, not always one (M-5) by @rburns629 in #120
  • fix(cli): a catalog release pins each product to the signed image of its own commit, and :latest moves only once signed (T-6) by @rburns629 in #123
  • fix(console): the first admin's password is never printed, so no log keeps it (L-2) by @rburns629 in #124
  • fix(ci): the release scans what it pushes, can't half-publish a version, and only runs on main's tags (X-1, X-5, X-6) by @rburns629 in #125
  • fix(catalog-template): the gcp samples install pyasn1 0.6.5 or later by @rburns629 in #134
  • fix(catalog-template): gitleaks and Semgrep gate publishing, and the README says which checks main's ruleset requires (T-5) by @rburns629 in #126
  • test(cli): the fake tofu answers a bare state list, as Forget now asks by @rburns629 in #135
  • fix(console): each person may ask the chat agent 30 questions an hour and 200 a day, and bots none (D-5) by @rburns629 in #115
  • fix(cli): the AWS catalog grant compares the reader's role before changing it, and leaves another reader's alone unless --replace (M-4) by @rburns629 in #122
  • fix(catalog-template): a retried create waits for what it found to be ready (T-3) by @rburns629 in #130
  • fix(catalog-template): s3-bucket is tagged as it's made, and a retry finishes its setup (T-4) by @rburns629 in #131
  • fix(catalog-template): app-hosting deploys the team's images, as the team's runtime account (T-7) by @rburns629 in #132
  • fix(cli): every chart vertra installs is checked by its SHA-256 first, and kind's registry runs by digest (L-4) by @rburns629 in #127
  • fix(ci): Go code gets Semgrep's Go rules and govulncheck (X-2) by @rburns629 in #128
  • fix(kernel,console,catalog-template): a sample's cleanup deletes only what it made for the team its run is for (T-2) by @rburns629 in #133
  • test(console): the stuck-run test starts the recorder's cursor from the oldest by @rburns629 in #136
  • docs: the v* tag ruleset is in place, and vertra checks the charts it installs by @rburns629 in #137
  • fix(console): the database pool is at least two connections by @rburns629 in #138
  • fix(cli): the console's AWS audit copy is create-only by IAM, not only by its own request (M-6) by @rburns629 in #139
  • fix(cli): Secrets vertra writes are applied server-side, so no plaintext copy lands in their last-applied annotation (L-7) by @rburns629 in #146
  • fix(cli): the cost screen leaves out the Postgres Pod a managed database doesn't run, and names what it doesn't price (M-8) by @rburns629 in #150
  • fix(console): a check that couldn't start shows a plain sentence, not the platform's error (C-8) by @rburns629 in #140
  • fix(console): a scan's verdict is recorded only by the reconcile whose Job it still is, and a failed one outlives its image (D-8, D-9) by @rburns629 in #142
  • fix(cli): a catalog's schemas resolve only references within themselves, never a file or URL (L-10) by @rburns629 in #148
  • fix(helm): the gateway's image can be pinned by digest, as the console's can (H-14) by @rburns629 in #153
  • fix(cli): no home directory means no tools directory, not a shared one, and downloads' checks are tested (L-12) by @rburns629 in #154
  • fix(gateway): the last traffic counts survive a rollout, settings of 0 are refused, and comments say what it writes (G-15, G-16, G-19) by @rburns629 in #157
  • fix(make): make test runs the gateway's tests, quickstart restarts the gateway, the kind context is always kind's, and two GKE targets work (X-8, X-9, X-10, X-11) by @rburns629 in #158
  • fix(console): a call with no one signed in is logged, not written to the audit log (C-10) by @rburns629 in #156
  • fix(perf): a cloud cluster is held to the vertra-perf prefix, as kind's is (X-7) by @rburns629 in #144
  • fix(helm): kernel-api admits the console's own Pods, not the gateway, and the gateway mints no token in a retired namespace (G-14, G-20) by @rburns629 in #172
  • fix(console): a secret can't switch between pasted and synced without a revoke (C-13) by @rburns629 in #169
  • docs(kernel): the metrics table lists kernel_unschedulable_steps, and the claim trace the product cap (K-13) by @rburns629 in #174
  • fix(console): single sign-on's settings and the memberships its groups change are audited (D-12) by @rburns629 in #178
  • fix(ci): only the highest version released moves latest and is pinned in the template (X-4) by @rburns629 in #141
  • fix(gateway): callers are looked up by uuid, a retired team calls as nobody, and API tokens' use is kept (G-9, G-12, G-17, G-18) by @rburns629 in #143
  • fix(cli): apply stops when Google sign-in's saved callback isn't where the console is now (M-7) by @rburns629 in #145
  • fix(helm): the kernel's migration Secret goes once the hooks succeed, and the by-hand commands are complete (H-13, H-15) by @rburns629 in #151
  • fix(cli): an unreached aws destroy offers to delete only the volumes its cluster's tags name (M-9) by @rburns629 in #155
  • fix(cli): the AWS backups bucket is versioned, so its writer can't destroy an earlier dump (L-8) by @rburns629 in #159
  • fix(cli): init's budget offer tells unread budgets from none, and a failed create keeps the answers (M-10) by @rburns629 in #160
  • fix(kernel): timing settings that don't work together stop it at startup (K-10) by @rburns629 in #166
  • docs: the architecture figures draw the gateway, and AGENTS.md names every example (X-12) by @rburns629 in #167
  • fix(console): a schedule that can't run again says why, and a time claimed but not started is started by another console (D-11) by @rburns629 in #170
  • fix(cli): destroy on an existing cluster deletes only the team namespaces the console made, and offers to uninstall the add-ons vertra installed (L-5, L-13) by @rburns629 in #171
  • fix(kernel): the refusals list includes RESOURCES_NOT_ALLOWED, and a step can't claim a refusal as its own failure (K-12) by @rburns629 in #173
  • test(console): the rare-day cron test's time bound holds under load by @rburns629 in #177
  • fix(console): the audit copy is compared field by field, so a new audit_log column isn't a false alarm (D-13) by @rburns629 in #181
  • fix(console): a guide that failed to load doesn't stick to the Docs page (F-2) by @rburns629 in #182
  • fix(catalog-template): the cloud samples install exactly the packages they pin, by hash (T-8) by @rburns629 in #187
  • fix(console): a short JWT_SECRET stops startup, and keyless bots need a real audience (C-14) by @rburns629 in #149
  • fix(kernel): the watch forgets changes to Jobs no one polls here any more (K-7) by @rburns629 in #163
  • fix(console): Home's Coming up shows each run in its schedule's own time zone, yours on hover (F-5) by @rburns629 in #179
  • fix(console): signing in goes back to the linked page, a restarting console isn't a sign-out, and signing out forgets the cart (F-8, F-9, F-10) by @rburns629 in #183
  • fix(console): a failed load says so, with Try again, instead of showing an empty list (F-6) by @rburns629 in #185
  • fix(cli,catalog-template): no team may be called bases, catalog-release, vertra-system or kube-* (T-9) by @rburns629 in #190
  • fix(catalog-template): gcp/project finds the project another step made a moment ago (T-11) by @rburns629 in #191
  • docs(catalog-template): what the template says matches what it does (T-13) by @rburns629 in #192
  • fix(gateway): nothing of an API is told before sign-in, and what callers claim about themselves stops at the gateway (G-10, G-11, G-13, G-20) by @rburns629 in #152
  • fix(console): an API's page lists the gateway's request steps in their new order (G-10, G-11) by @rburns629 in #194
  • fix(console): a multi-platform image is scanned for every platform the cluster runs (D-10) by @rburns629 in #161
  • fix(cli): the gcp module's Cloud SQL roles hold for its own instance only (L-9) by @rburns629 in #165
  • feat(console): the Admin page sets how many questions each person may ask the chat agent (D-5) by @rburns629 in #175
  • fix(console): an email is one account however it's typed (C-9) by @rburns629 in #176
  • fix(helm): the console's default deny may carry no egress rule, and no-metadata must except the metadata server (H-12, H-16) by @rburns629 in #180
  • fix(catalog-template): the local samples refuse a result too big for result.json as TEXT_TOO_LONG (T-12) by @rburns629 in #193
  • fix(catalog-template): a new catalog ignores what local test runs and builds leave behind (T-14) by @rburns629 in #189
  • fix(catalog-template): code scanning covers every language the starters offer (T-10) by @rburns629 in #188
  • fix(console): a watch checks the caller's credential again on every read (C-12) by @rburns629 in #164
  • fix(kernel): database-setup lets the console pass on only what it gives the gateway (K-11) by @rburns629 in #168
  • refactor(console): one errorMessage and one catalog-file link builder in the frontend (F-11) by @rburns629 in #184
  • fix(console): a pasted secret's value is masked, and never spell-checked or remembered (F-7) by @rburns629 in #186
  • fix(kernel): a step leaves RUNNING only once it's known to have stopped, and only the kernel's own Job is adopted (K-4, K-5, K-6, K-8, K-9, K-14) by @rburns629 in #162
  • docs: review follow-ups (who sees a hidden API, what a cookie's path keeps apart) by @rburns629 in #197
  • feat(gateway): behind trusted proxies, a backend sees the caller's address (G-10 follow-up) by @rburns629 in #199
  • fix(kernel): a step's Job starts suspended and is resumed only once its request is the kernel's (K-8 follow-up) by @rburns629 in #198
  • build(deps): Bump the base-images group across 4 directories with 1 update by @dependabot[bot] in #195
  • build(deps): Bump the console-npm group across 2 directories with 11 updates by @dependabot[bot] in #196
  • chore: what the slop sweep found: stale references, dead code, a quiet failure by @rburns629 in #200
  • docs: TODO.md holds only open work again by @rburns629 in #201
  • fix(cli): vertra apply sets an AWS catalog's OIDC subject before the publish role needs it; the catalog guide catches up by @rburns629 in #202

Full Changelog: v0.8.0...v0.9.0