

Seven sites publish inside the transaction
Connection held while SNS decides
Phantom event if the commit falls over
Consumer reads the rows before they arrive
Read the page, then read the node_modules
afterCommit hooks run in a finally
Nested commit is a no-op,
mate
The cheap path was never that cheap to me
Commit the row,
send it later
One statement, SKIP LOCKED,
no transaction spanning
Commit the row,
send it later
Every minute the publisher comes
Claim,
publish,
mark, sweep
Commit the row, send it later
Worktree off develop, yarn install immutable
Migration, model,
OutboxManager
ULIDs minted in the same millisecond
Came back shuffled, so make them monotonic
RETURNING gives you rows in any order
Wrap it in a CTE,
ORDER BY id
Permission denied on the DELETE
The api role never had the grant,
you see
Commit the row, send it later
One statement, SKIP LOCKED,
no transaction spanning
Commit the row, send it later
Attempts plus one at the claim,
not the fail
Twenty strikes
and the run throws
Commit the row, send it later
Four referral tests went red.
readAt was null.
The local server shims
SNS on the prototype,
and my handler grabbed the client first.
Load it lazy.
Green.
Then: "what's in those two files you gobbled up with the commit?"
ablyAuth. alivecorAuth. Your staged work,
not mine.
Soft reset.
Unstage.
Recommit. Restage.
Never again.
Written down.
"What are you on about,
lambda_functions?"
Grep wider, it's in variables.sh
Derived from list-functions at apply time
So the alarm shows up after the deploy
Two grants files and they already drift
provision.sql for the real environments Locally the cron shim runs it every
minute
Same as the Lepu poller,
nothing new
Commit the row, send it later
The action succeeds and the event catches up
Commit the row, send it later
That's the whole point,
not a bug
Draft PR,
mermaid in the body
Commit the row, send it later
Nudge not built.
Decide on sandbox.
Four direct sites still publish inside.
Doing them now.
Opus is still reading.
Commit the row.
Send it later.