Start-up recovery and the 60-second sweep were two parallel sets of
functions doing the same job. recoverDeliveries now does it, run once at
start and then on every tick from one goroutine. A retrying delivery is
scheduled after what is left of its backoff, now if nothing is left, so
backoffElapsed is gone. The run at start still takes every pending
delivery and opens a webhook database whose file is missing; a tick
still takes only pending deliveries untouched for 15 minutes, at most
500, and skips a missing file. A retrying delivery is taken before its
event is read. The README describes one routine.
Model: opus-5-5
The SQLite archive files the database target writes are now managed by
internal/archive: file names, rotation, the writer of each target,
pruning, renames, the gzipped JSON export, the idle sweep, and the
rename lock, which was Handlers.renameMu. The database target stays in
internal/delivery and writes through archive.Writers; the engine still
closes the writers when it stops, so the stop order is unchanged. The
handlers take the rename lock from archive.Archives.RenameLock, and
rename and evict through archive.Archives.
Exported names drop their Archive prefix in the new package. The tests
moved with the code; the two that check a delivery's recorded attempt
stay in delivery.
Model: opus-5-5
Target.MaxRetries (max_retries) becomes Target.MaxAttempts
(max_attempts), and the form field, parser and ceiling are renamed to
match. The column is NOT NULL and a check constraint refuses a value
below 1, so every target, whatever its type, is stored with at least
1. 1 is a single attempt with no circuit breaker; above 1 retries
behind the breaker up to that many attempts. Both target forms refuse
0 and less; a blank field on the add target form gives 1, and on the
edit form keeps the stored value. The target page shows the plain
number, and the README's two paragraphs on 0 against 1 become one
sentence.
Model: opus-5-5
finished_at, archived_at and the updated_at stamp a re-dispatch
writes came from a bare time.Now(). The SQLite driver stores a time
as time.Time.String(), which appends the monotonic clock reading
(" m=+…") to a time read from time.Now(). All three now come from
GORM's own clock, NowFunc(), which stamps created_at, so they are
stored in the same form. The tests for each write read the stored
text back.
Model: opus-5-5
Three tests failed at random on a busy host. The browser test now waits for each page a click opens to load, and for Alpine.js to start on it, before reading it. The delivery tests' drain takes what is already queued instead of racing a 25 ms timer, since both dispatch paths queue before they return. The test phase keeps the tests' temporary directories on a tmpfs, because SQLite waiting for the disk made `internal/handlers` slow under load; the timeout and the parallel cap are unchanged.
Deviation: the delivery tests do not wait with a deadline; dispatch has finished when they read.
Unverified: `internal/handlers` at a host load of 260 to 290.
Model: opus-5-5
Restart recovery could find a just-written delivery pending, send it
and release it before the receiver's Notify queued the same delivery.
Notify's claim then succeeded on the released id, and the worker sent
it again because the new-task path never read the delivery's row.
Before sending a new task the worker now reads the delivery's status
by primary key and skips the task unless the row still says pending,
as the retry path already does for retrying. Nothing else can change
the row while the worker owns the delivery. A row left pending by a
failed bookkeeping write is still sent again.
loadRetryDelivery is renamed loadDelivery now that both paths use it.
Model: opus-5-5
clawbot