Skip to content

fix(0286): read the post-0304 amm summary in the re-ingest script - #350

Merged
karczuRF merged 4 commits into
developfrom
fix/0286_reingest-amm-step-after-0304
Sep 24, 2026
Merged

karczuRF merged 4 commits into
developfrom
fix/0286_reingest-amm-step-after-0304

Conversation

@karczuRF

Copy link
Copy Markdown
Collaborator

Summary

  • The re-ingest script's AMM step read events with no apply order:, a line task 0304 removed from events-backfill, so the first Soroban-era month (202402) would crash in Python after the AMM candles were written. It now reads negative apply order: and swaps dropped (unresolved):, after the --dry-run pass as well, and stops before the write if either line is missing.
  • --amm ssh records dropped swaps like --amm mtls did. Both modes add a note for events with a negative apply order.
  • The --amm mode is checked in preflight and in each month's gates step, before the snapshot and the DROP. --amm ssh without --ssh stops there.
  • The default is now --amm ssh, because events-backfill on develop has no --transport. The default password is asked for only when a Soroban-era month is still to do.
  • Runbook: the cleanup rule runs at 03:00 UTC, not 02:00.

Stage A (running with --amm stop to 202401) is not affected. The Comet gate was already added under 0300.

The AMM step parsed 'events with no apply order:', which task 0304
removed, so the first Soroban-era month would crash after events-backfill
had written. Read 'negative apply order:' after the dry-run pass too, and
check the --amm mode before the snapshot and DROP. Default to --amm ssh:
events-backfill on develop has no --transport.
karczuRF added a commit that referenced this pull request Sep 24, 2026
Records PR #350, the git pull after stage A, the --ssh target, the host
events-backfill build, --discover-pools and the stage B command.
A missing or non-executable events-backfill binary raised OSError out of
require_transport_flag, which neither the preflight gates nor main()
catch: the operator got a traceback instead of the failed-gates list.
Raise Stop instead, so preflight and each month's gates step report it.
The mock's 400 ms delay against a 350 ms bound left under 300 ms of
slack, and the clock also covered building the SDK client. A loaded
CI runner took 587 ms and failed a deadline that did cut the call off.
Build the router before timing, and set the delay to 3 s against a 2 s
bound: only a response that did not wait for the mock can pass.
@karczuRF
karczuRF merged commit af716df into develop Sep 24, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant