Comparison to Rails¶
firm is a faithful port of the Rails Solid stack — queue, cache, and pub/sub — not a reinterpretation. Where it differs, it's deliberate and documented.
What's the same¶
- Schema. Every column, index, and the execution-table lifecycle match the gems one-for-one.
The tables carry firm's own names (
firm_*). - Behavior. The job lifecycle, the four process roles, concurrency semaphores with the
expires_atfailsafe, recurring(task_key, run_at)dedupe, FIFO cache eviction with sampling size estimation, thekey_hashlookup, the probabilistic expiry trigger, and the pub/sub polling listener with per-channel cursors,channel_hashrouting, and probabilistic message trimming — all match. - Locking.
FOR UPDATE SKIP LOCKEDon PostgreSQL/MySQL, exactly as the gems (the queue's claim path, and the pub/sub trim).
Deliberate divergences¶
| Area | solid_queue / solid_cache / solid_cable | firm |
|---|---|---|
| Retry counting | tracked by Active Job | a jobs.attempts column (we own retries) |
| Crash recovery | orphaned claims are marked failed | orphaned claims are re-readied (at-least-once; another worker finishes them) |
| SQLite concurrency | row-level locking tests are skipped | BEGIN IMMEDIATE gives the same guarantee, so concurrency is fully tested on SQLite |
| Recurring schedules | Fugit (cron and natural language), timezone-aware | cron only (croniter), evaluated in UTC |
| Job arguments | Active Job + GlobalID (pass records) | JSON + datetime/date/Decimal/UUID (pass IDs) |
| Pub/sub trimming | inline TrimJob per broadcast |
async probabilistic trim on a background thread (+ manual trim() / CLI) |
| Pub/sub delivery | Action Cable adapter | a standalone Channel with broadcast/subscribe/unsubscribe (no Action Cable) |
| Cache value coder | Marshal (arbitrary Ruby objects) | JSON by default (safer against a writable cache table); PickleCoder is one import away |
| Cache entry expiry | per-entry expires_in / expires_at |
global expiry only — max_age + max_size FIFO trim; no per-entry TTL (planned, see ROADMAP.md) |
| Cache failure safety | a dead cache DB degrades reads and writes | reads degrade to a miss (routed to on_error); writes still raise — a silently dropped write is a worse surprise than an error |
The at-least-once recovery choice means jobs should be idempotent — see Retries & failures.
Recurring schedule cron expressions are evaluated in UTC: croniter runs over firm's
naive-UTC clock, whereas solid_queue's fugit schedules honor the application's configured time
zone. Porting a schedule like 0 9 * * * therefore shifts its fire time by your UTC offset — it
fires at 09:00 UTC, not 09:00 local. Adjust the cron fields to UTC when migrating.
Beyond the port: firm-audit¶
firm.audit is not a port — there is no solid_audit gem in Rails. It's an original firm
module: an append-only, database-backed audit log that shares the ported modules' "you already
have a database" thesis, but isn't a reproduction of anything in Solid. See
firm-audit overview.
Not (yet) ported¶
firm stands alone — there's no Rails — so the Active Job / Action Cable ecosystems aren't reproduced, and a few pieces are still future work:
- Active Job integration (callbacks,
perform_later, GlobalID, middleware) — replaced by the standalone@jobdecorator. - Action Cable integration (channels, connections, the websocket server) —
firm-channelports the database-backed pub/sub broker, not the websocket layer on top of it. - Active Support / Notifications instrumentation — replaced by lifecycle hooks + your own logging.
- Cache sharding (multi-database consistent hashing) and the Mission Control dashboard.
- Continuations (
ActiveJob::Continuable) — checkpointing inside a job and resuming after an interruption. A recovered job restarts from the top, which is why jobs must be idempotent.
See ROADMAP.md in the repo root for the roadmap — the six things planned next. Everything
else lives in the issue tracker.