7 Commits

Include renames

Author	SHA1	Message	Date
Jonathan KaiserandClaude Sonnet 5	9906dd7b09	fix: correct test setup bugs found by actually running the suite ... Set up a working Docker-based Rails environment (this session's shell had no compatible Ruby/Bundler) and ran the full suite against PR #3298. Two real bugs surfaced that static checks couldn't have caught: - WorkerAiHealth::Snapshot.new(**{...}.merge(overrides)) needs the double-splat -- a bare Hash isn't auto-converted to keyword arguments. Both test snapshot builders passed a positional Hash instead, which raised "missing keywords" for every field on every call. - assert_enqueued_with/assert_no_enqueued_jobs need `include ActiveJob::TestHelper` explicitly in a plain ActiveSupport::TestCase -- every other model test in this codebase that uses them does the same; I'd wrongly assumed it was available process-wide. With both fixed: 7510 runs, 29877 assertions, 0 failures, 0 errors, 30 skips for the full suite; rubocop, erb_lint, and brakeman all clean. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A9494Pxh4LZKnNLTGfFXPw	2026-09-02 07:15:54 +02:00
Jonathan KaiserandClaude Sonnet 5	b9b5a7c58b	feat: verify AI configuration and provider liveness from worker processes ... Closes #3169. The AI status page (#3145, PR #3155) proves only that the `web` process resolved a valid-looking configuration and can reach the configured provider from its own network context. Most AI workloads -- assistant responses, PDF processing, embeddings, auto-categorization, and merchant detection -- actually run in Sidekiq `worker` processes, which can differ from `web` in environment, DNS, proxy rules, network policy, or even loaded credentials (workload-specific overrides, an updated Secret without a pod restart, `web` recreated without `worker`). A passing web check says nothing about whether a worker can do the same. ## What this adds `WorkerAiHealthCheckJob`, queued on demand from a new "Verify worker configuration" button on System health -> AI status. It runs the same bounded, non-destructive probes `AiHealth` already runs, but from inside whichever Sidekiq worker process dequeues it, and records the result via `WorkerAiHealth`: process identity (hostname:pid), checked-at time, a non-secret configuration fingerprint (effective provider, model, redacted endpoint, vector-store adapter/embedding config), and probe outcomes. The AI status tab lists every recorded result -- most recent first, kept for `WorkerAiHealth::RETENTION` (15 minutes) -- each labeled with a status pill (`Passing` / `Failing` / `Stale`, the last once older than `STALE_AFTER`) and a configuration pill comparing it against the web snapshot (`Matches web` / `Differs from web`), with a failure-reason list reusing the existing failure-code translations when a probe failed. ## Implementation constraints from the issue, addressed directly - **Cannot reuse a web-cached probe result, or vice versa.** `AiHealth.new` gained an injectable `probe_cache:` (default `Rails.cache`, matching today's behavior). The worker job passes a fresh `ActiveSupport::Cache::NullStore` instead, so every worker check is a live call that neither reads a web-cached entry nor leaves one behind. - **A single job only verifies one worker.** Documented on the button (`coverage_notice`) and in the docs: with multiple replicas, a passing result names one process, not the fleet. Queuing again samples another. - **Never persists or displays a raw credential.** `WorkerAiHealth::Snapshot` only carries redacted endpoints (AiHealth already redacts these before they reach the job) and provider/model/status fields -- there is no field for a token to occupy. A structural test asserts this stays true. - **Failures land in both places an operator already checks.** Same destinations as `AiHealth::Probe`'s own failures: `Rails.logger` and `DebugLogEntry` (new `ai_health_worker` category), tagged with the process identity. - **Results carry a clear status**, including the `pending` case implicitly (no result yet renders an explanatory empty state) and `stale` for a result whose process may no longer reflect current state. - **DB-backed vs ENV-backed settings are labeled.** A new info block next to the worker results explains which UI settings propagate automatically (rails-settings-cached invalidates the shared cache on write) versus which require restarting/recreating both `web` and `worker`. ## What this deliberately doesn't do Full-fleet coverage (every process publishing a periodic fingerprint) -- the issue lists this under "Other options to consider," not the acceptance criteria, and Sidekiq's normal dispatch doesn't target every process without an explicit per-process coordination mechanism. This PR implements the on-demand, single-check design the acceptance criteria actually describes ("An administrator can request an asynchronous worker-side check", "does not imply full-fleet coverage"); periodic fleet-wide publishing is a natural follow-up if operators need it. ## Testing - `WorkerAiHealth`: recording/reading, same-process replacement vs. cross-process coexistence, MAX_RESULTS bounding, staleness, status derivation (failure codes / component statuses / function-calling refusal), `matches_web?` comparison, and the credential-field structural guard. - `WorkerAiHealthCheckJob`: records a passing/failing snapshot naming this process, writes failures to Rails.logger + DebugLogEntry (and only on failure), never leaks the access token, and -- the defining property -- is proven to construct `AiHealth.new` with an isolated `NullStore` rather than the shared web-facing cache. - `Admin::SystemHealthController`: empty state, a rendered result with matching/mismatched configuration, a failing result's failure reason, a stale result, the `verify_worker_ai` action enqueuing the job and redirecting with a flash notice, and that non-super-admins and unauthenticated requests cannot trigger it. I could not run the Rails test suite in this environment (no working Ruby/Bundler toolchain available locally -- Ruby 2.6 system Ruby vs. the project's required 3.4.9, no way to install without sudo/Docker access). Every file was checked with `ruby -c`, YAML files with `YAML.load_file`, and the ERB view with `ERB.new(...).src`, plus careful manual tracing of each test against the production code paths it exercises, but CI should be treated as the first real run of this suite per the repository's own guidance for exactly this situation. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01A9494Pxh4LZKnNLTGfFXPw	2026-09-02 07:15:54 +02:00
3 people	f7a7736e1b	Surface probe timeout separately from LLM request timeout on System Health (#3276) ... * Surface probe timeout separately from LLM request timeout in admin System Health - Expose probe_request_timeout in AiHealth (mirrors Probe#timeout) - Split the ambiguous 'Request timeout' row into 'LLM request timeout' and a new 'Health-check probe timeout' row - Forward AI_HEALTH_PROBE_TIMEOUT / AI_HEALTH_PROBE_CACHE_TTL in compose.example.yml and compose.example.ai.yml - Remove now-orphaned labels.request_timeout locale key (i18n-tasks) - Add regression test asserting both values surface distinctly * Address PR #3276 review feedback - Restore the credential-redaction assertion on the PDF probe status test to match the token the test actually configures (it had been changed to a placeholder that appears in no code path, making the assertion a no-op and silently removing its coverage). Sanitize the new distinct-timeouts test to a non-secret token and assert it is absent. - Consolidate the probe timeout into a single public AiHealth::Probe.timeout class method and have AiHealth#probe_request_timeout_value delegate to it so the two can never drift (rather than re-implementing the same ENV.fetch + positive-check logic). - Add unit coverage for Probe.timeout exercising the env var and the fallback path for missing/zero/negative/non-numeric values. --------- Signed-off-by: Juan José Mata <juanjo.mata@gmail.com> Co-authored-by: jaysbeekay <jaysbeekay@users.noreply.github.com> Co-authored-by: Juan José Mata <juanjo.mata@gmail.com>	2026-09-01 01:48:05 +02:00
Juan José MataandClaude	f78303ebbe	Add function-calling probe to detect tool-use support (#3255) ... * feat(ai-health): name the missing function calling behind an opaque chat error The assistant reads accounts, transactions and holdings through function calls, so every chat request carries a `tools` payload. A model without function-calling support rejects it — OpenRouter answers a bare 404 — and the operator sees only that status code, with nothing pointing at the model. Both earlier attempts at this guessed from the chat-time error; the AI status page already runs live probes, so let it answer the question directly instead. `AiHealth::Probe#function_calling` asks the configured model for one trivial tool call the way the assistant asks for its own: chat completions with `tools` for OpenAI-compatible endpoints, the Responses API for hosted OpenAI, and `messages.create` with `tools` for Anthropic, carrying the same strict schema `Provider::Openai` sends. Reading it against the plain LLM probe is what makes the verdict sound rather than a guess at 404s: plain chat passing while the same request with tools fails means the model has no function calling; a response that carries no tool call means the endpoint took the tools but the model ignored them; both failing, or a timeout, stays an ordinary probe failure. The AI status card gains a Function calling (tools) row, an alert naming the fix for each of the two bad outcomes, and a failure reason. The hosting settings model field now says the assistant needs a tools-capable model and links super admins to the check. Refs #830 Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DWCmRQ1ry26JnhA79s1ZKH * fix(ai-health): only call a refusal a refusal, and probe the route chat takes Two findings from review of the function-calling probe. A tools request can fail for reasons that say nothing about tool support: a 429, a 500, a dropped connection, an unreadable body. Reading any non-timeout failure as `:unsupported` sent the operator hunting for a new model over a transient blip. Only a 4xx the service chose to answer with — excluding the ones that mean "not now" or "not you" — is a refusal of the tools payload; everything else stays an ordinary probe failure. The bare 404 from OpenRouter that this page exists to explain still reads as missing function calling. `Provider::Openai#supports_responses_endpoint?` is the real routing decision and `OPENAI_SUPPORTS_RESPONSES_ENDPOINT` can flip it either way, so choosing the API from "is the endpoint custom" could probe Chat Completions while chat uses Responses, or the reverse — reporting on a path the assistant never takes. Ask the provider instead, and cache the two routes under separate keys. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DWCmRQ1ry26JnhA79s1ZKH * fix(ai-health): confirm a tools refusal before blaming the model A client error on the tools request can mean "your tools payload" or "your request, tools or not" — an invalid schema, a route the endpoint does not serve, a model it will not run. Splitting those on the status code alone still put a 422 from an endpoint contract on the model's account and told the operator to go find another one. The probe now confirms it: when the tools request comes back a client error, it asks again with the tools taken off. Only if that lands is the tools payload what was turned down, and the probe says so with its own failure code — provider-agnostic, and no reading of error text for the word "tool", which would only ever fit the provider it was written against. Statuses that mean "not now" or "not you" (401, 402, 403, 408, 429) never get a second ask. `AiHealth` now just reports the probe's verdict instead of inferring one from the status. The troubleshooting fix no longer points at an OpenRouter free tier: those providers commonly log prompts and completions for training, and every assistant tool call carries accounts, transactions, and holdings. It points at the model recommendations already in this doc, and says why free tiers are the wrong place to look. Co-Authored-By: Claude <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01DWCmRQ1ry26JnhA79s1ZKH --------- Co-authored-by: Claude <noreply@anthropic.com>	2026-08-31 19:47:48 +02:00
Juan José Mata	2abce5f5b7	Verify PDF support with a synthetic health probe (#3191) ... * Add synthetic PDF health checks * Require exact marker in PDF health probes * Report PDF health paths separately * Refactor application code * Remove unrelated schema dump changes * Simplify synthetic PDF validation --------- Signed-off-by: Juan José Mata <juanjo.mata@gmail.com>	2026-08-27 00:14:50 +02:00
Juan José Mata	fd6f4ff078	Add live AI checks to system health (#3155) ... * Add live AI checks to system health Give super admins a dedicated AI status view with bounded liveness probes for LLMs, vector stores, pgvector, and embedding endpoints. Record sanitized failures in both the system debug log and Rails logger, and document the recommended local configuration.\n\nCloses #3145 * Fix AI health CI checks * Address AI health review feedback * Correct Ollama model preload guidance * Distinguish OpenAI-compatible providers * Make Ollama startup readiness explicit * Recognize Cloudflare AI endpoints	2026-08-24 22:41:08 +02:00
3 people	6439a731ab	feat(self-hosting): surface Sidekiq-unhealthy nudge + admin system health page (#1906) ... * feat(self-hosting): surface Sidekiq-unhealthy nudge + admin system health page (#1481) When the Sidekiq worker container isn't running — the most common Docker Compose misconfiguration in self-hosted setups — every background job silently never executes. Balance calculations, net-worth updates, and account syncs stall. The UI shows zeros and "No balance data available for this date" without explaining why (#1481, #1047). Per jjmata's resolution on the issue, this PR ships both halves of the fix in one pass: 1. A user-facing nudge banner that appears on every authenticated page when Sidekiq isn't processing jobs. Tells the user their data may be stale; doesn't pretend zeros are real. 2. An admin-only deep link from that banner into a new `/settings/admin/system_health` page (super-admin gated, matching the existing admin namespace contract) showing live Sidekiq state: process count, last heartbeat, max queue latency, job counters, and per-queue depth. ## What changed - New `SidekiqHealth` PORO (`app/models/sidekiq_health.rb`) eagerly loads ProcessSet + Queue + Stats in one pass and exposes `healthy?` plus a stable `reason` symbol (`:redis_unreachable`, `:no_worker_processes`, `:stale_heartbeat`, `:queue_backed_up`). Any Redis/Sidekiq failure during the eager load is caught and surfaced as `:redis_unreachable` so a degraded broker never crashes the layout. - `ApplicationController#current_sidekiq_health` memoizes a single instance per request via `helper_method` so the layout, banner partial, and any controller checks share one Redis round-trip. - New `app/views/shared/_sidekiq_health_banner.html.erb` rendered from `_htmldoc.html.erb` when `Current.user` is present and the health check is failing. Banner shows the user-facing message to everyone; the "View system health" CTA + reason detail are gated on `Current.user&.super_admin?`. - New `Admin::SystemHealthController#show` (inherits the existing `Admin::BaseController`, so super-admin gating is enforced for free) + view rendering status, counters, and per-queue breakdown. - Routes: `resource :system_health, only: :show` inside the existing `namespace :admin`. - Settings nav: new "System health" entry under the Advanced section, gated on `super_admin?` to match `sso_providers_label` and `users_label`. - i18n: new `shared.sidekiq_health_banner.*` keys (title, body, CTA, per-reason explanations) and a full `admin.system_health.show.*` namespace for the new admin page. English-only, matching how `ds.pill.*` and other DS keys are scoped. ## Why - jjmata: "Let's take both approaches ... a nudge about 'data unavailable' which hyperlinks to the admin UI if you are an admin only (not for other types of users) sounds like the best path forward. **Any takers for the PR?**" (#1481) - smurfpandey: "We can add a section in Settings for superadmins to see 'health' of the application/host." - The detection signal is conservative on purpose: - `PROCESS_HEARTBEAT_TIMEOUT = 2.minutes` tolerates deploy restarts and brief Redis blips without flapping. - `LATENCY_THRESHOLD = 5.minutes` is well above the sync-job tail under default `config/sidekiq.yml` concurrency. ## Validation This worktree runs on Windows without a local Ruby toolchain, so I could not run `bin/rubocop`, `bundle exec erb_lint`, `bin/brakeman`, or `bin/rails test` locally. CI will run the full matrix on the PR: - `lint` — `bin/rubocop -f github` - `lint_js` — `npm run lint` (no JS touched, should be green) - `scan_ruby` — `bin/brakeman --no-pager` - `scan_js` — `bin/importmap audit` - `test_unit` — `bin/rails test` (includes 7 new tests under `test/models/sidekiq_health_test.rb` and 4 new under `test/controllers/admin/system_health_controller_test.rb`) - `test_system` — `DISABLE_PARALLELIZATION=true bin/rails test:system` - `pipelock` — secret + agent-security diff scan Manual checks done in this worktree: - Re-read `CONTRIBUTING.md` and `.cursor/rules/project-conventions.mdc`. PORO under `app/models/` per Convention 2. No new gem dependency per Convention 1. Banner uses semantic tokens (`bg-warning/10`, `text-warning`) per the design-system rules. No `lucide_icon` direct call — uses the `icon` helper per CLAUDE.md. - Confirmed `Sidekiq::ProcessSet` / `Sidekiq::Queue` / `Sidekiq::Stats` are the same APIs Sidekiq 7+ exposes (we're on Sidekiq 8.x per the `Gemfile.lock` comment in `config/initializers/sidekiq.rb`). - Tests stub `Sidekiq::ProcessSet.new` / `Sidekiq::Queue.all` / `Sidekiq::Stats.new` so the suite doesn't need Redis populated. - The admin route lives inside the existing `namespace :admin` so `Admin::BaseController#require_super_admin!` enforces auth — no new authorization surface added. ## Notes - No public API endpoints, no rswag specs, no OpenAPI changes. - No migrations, no model changes outside the new PORO. - No background jobs touched. - English-only locale entry, mirroring the `ds.*` / `admin.invitations.*` precedent in this repo. Other locales fall back to English. - Detection thresholds are constants on `SidekiqHealth` so they're easy to tune from a follow-up PR if the defaults turn out to flap on any real-world deployment. - The banner positions itself at `top-20` (below the impersonation / super-admin bars) and uses `z-40` (below the `z-50` notification tray). Single-screen overlap with mobile flash toasts is acceptable for V1. Refs: #1481, #1047 * fix(self-hosting): address review on Sidekiq health PR (#1481) - `Admin::SystemHealthController#show` now reads from the request-memoized `current_sidekiq_health` instead of building a fresh `SidekiqHealth.new`, so the controller and the layout banner share one Redis round-trip. - `SidekiqHealth#reason` now treats `last_heartbeat_at.nil?` the same as a stale beat: a registered process that hasn't published a heartbeat is not "healthy". Previously the check short-circuited on the nil guard and silently fell through to the queue-latency branch. Added a unit test covering the `ProcessSet` entry with `"beat" => nil` case. - Settings nav: switched the "System health" entry's icon from `activity` to `heart-pulse` so it no longer duplicates the LLM Usage icon. - Routes: dropped the redundant `controller: "system_health"` option from the `resource :system_health` declaration — Rails infers `Admin::SystemHealthController` from the namespace, matching the style of the sibling `:sso_providers`, `:users`, `:invitations`, and `:families` admin resources. * fix(self-hosting): scope + cache Sidekiq health, admin-only banner (#1481) Addresses the second round of maintainer review on the Sidekiq health PR. - Skip the check entirely in managed mode. `current_sidekiq_health` returns `nil` unless `Rails.application.config.app_mode.self_hosted?`, so authenticated requests in managed deployments add zero Redis round-trips for this feature. - Cache the snapshot across requests via `SidekiqHealth.current` (Rails.cache, TTL `CACHE_TTL` = 60s default, env-overridable). The per-request memoization on `ApplicationController` is preserved on top, so even back-to-back self-hosted pages share one fetch. - Make thresholds operator-tunable. `PROCESS_HEARTBEAT_TIMEOUT`, `LATENCY_THRESHOLD`, and the new `CACHE_TTL` read from `SIDEKIQ_HEALTH_HEARTBEAT_TIMEOUT`, `SIDEKIQ_HEALTH_LATENCY_THRESHOLD`, and `SIDEKIQ_HEALTH_CACHE_TTL` env vars (seconds), with the previous values as defaults. Comments now explain the tuning rationale. - Gate the banner on `Current.user&.super_admin?` at the layout level rather than rendering a vague warning to family members who can't act on it. The partial no longer carries an internal admin check since the call site does it; non-admins see nothing. - Replace the hard-coded `top-20` offset with a computed offset based on which impersonation bars are visible (`top-4` / `top-20` / `top-36`) so the banner doesn't collide with the super-admin or approval bars when both are stacked above it. - `Admin::SystemHealthController#show` now bypasses the cache (`SidekiqHealth.expire_cache!` + `SidekiqHealth.new`) so an operator who just restarted the worker sees fresh state instead of a stale 60-second snapshot. Also lets the page render in managed mode where `current_sidekiq_health` is nil. - Tests: add coverage for `.current` cache reuse and `.expire_cache!` forcing a re-query, swapping `Rails.cache` to a MemoryStore since the test env defaults to `:null_store`. * fix(self-hosting): route singular resource + drop assert_same on cached snapshot (#1481) Two CI failures surfaced once the full pipeline ran on this branch for the first time (it was gated on contributor approval until d04b78e): - Admin system-health controller tests returned 404. Singular `resource :system_health` in `config/routes.rb` makes Rails infer `Admin::SystemHealthsController` (it pluralizes the controller name even for singular resources), but the controller file is named `system_health_controller.rb` / `Admin::SystemHealthController`. Restore the explicit `controller: "system_health"` override that the previous "address review" commit dropped on the (mistaken) premise that Rails would infer it from the namespace — the sibling admin routes all use plural `resources` so they round-trip cleanly, this one doesn't. Comment now spells the gotcha out so the next reviewer doesn't try to "simplify" it again. - `SidekiqHealthTest#test_current_memoizes_across_calls_inside_the_cache_TTL` used `assert_same` on the two returns from `SidekiqHealth.current`. `ActiveSupport::Cache::MemoryStore` defaults to `dup_values: true` and Marshals on read, so a cache hit returns an `==`-equal but `equal?`-different instance. Replace the identity check with the behavioral assertion we actually care about: re-stub `ProcessSet` to raise on the second call, then assert the second `current` return is still healthy (proving Redis was not re-queried). * fix(i18n): drop redundant inline default on system_health nav label (#1481) `system_health_label` is already defined in config/locales/views/settings/en.yml, so the inline `default: "System health"` was a hard-coded English string in the template (DS Drift Patrol Rule 5). Use the bare locale lookup like the sibling nav entries. --------- Co-authored-by: John Baillie <johnbaillie2007@gmail.com> Co-authored-by: Khaostica <256858950+Khaostica@users.noreply.github.com>	2026-08-22 07:17:29 +02:00