[plainprogrammer]

A job's page at /admin/jobs/:id now counts the job's own run, so a job that failed reads 1 rather than "Attempts 0". The row is labelled "Attempts since queued".

Closes #61

Cause

BackgroundJobs::Entry#attempts returned executions from the job's stored Active Job envelope. Two things make that the wrong number:

• Active Job raises executions only in memory in perform_now (activejob 8.1.4 execution.rb:47). It is stored only on the new row a retry_on retry inserts, so a row's stored value is the runs before that row was queued and never its own run.
• An admin's retry calls FailedExecution#retry → Job#reset_execution_counters (solid_queue 1.7.0 job/retryable.rb:33), which sets it back to 0 on the same row. After that the row is identical to a freshly queued one.

So the queue's tables can't give a total across an admin retry.

What changed

• Entry#attempts is the stored executions plus one when the job is running, failed or finished. No new storage, and nothing is written to the queue's rows (ADR 0014).
• The page's label is "Attempts since queued".
• AC-6.5, contracts/api.md and the code comment say how the count behaves (spec 015 version 1.1.3). The research note that recorded "Attempts 0" points here.

State	Shows
Queued, waiting, scheduled (never run)	0
Running, failed or finished on its first run	1
Failed, then retried by an admin	0 while queued, 1 once it runs again
retry_on retries	1, 2, 3 across the rows; a waiting retry row shows the runs before it
Put back by a worker's graceful shutdown, then run again	1 → 0 → 1

Decision for you: the label

I chose the no-storage option and the label "Attempts since queued" during the work. The reviewer found that no single reading of "queued" is true in every row beside the page's own "Queued" time:

• After an admin retry the page can read "Attempts since queued 1 / Queued 03:15 / Failed 04:02" for a job that has run twice since 03:15, because a retry keeps the row's original created_at.
• A waiting retry_on row reads 1 although nothing has run since that row was queued.
• A job released by a graceful shutdown and started again reads 1.

Alternatives: keep "Attempts" and let AC-6.5 carry the caveats; remove the row; or add storage to count admin retries (a migration, and the primary and queue databases are separate files). Changing the label is one line in app/views/admin/jobs/show.html.erb plus the spec wording.

Verification

• bin/ci: passed. RSpec 1226 examples, 0 failures, 1 pending (the art agreement spec, unrelated).
• The new examples failed before the fix (4 model, 2 request). The retry_on examples run jobs through Solid Queue's real ClaimedExecution#perform. The graceful-shutdown example is a characterization that passed on first run; it fails under a mutation of attempts.
• An independent read-only review (Opus) confirmed the arithmetic against Solid Queue and Active Job's source and asked for wording fixes only, which the last four commits make.
• Not verified: a real worker in development (retry, shutdown), and the light theme. Dark-theme screenshots at 1280 px and in a 390 px frame showed the label on one line.

Notes

• The first row of a retry_on chain shows as Finished with 1 attempt although it raised; that is how Solid Queue records it and is unchanged.
• This and Catalog page reads "Running." after a crashed refresh #60's PR both edit spec 015's changelog; whichever merges second needs a small rebase. docs/roadmap.md still lists A retried job shows "Attempts 0" #61.

🤖 Generated with Claude Code