Skip to content

test: shorter index, resume, sv2, and mempool pads - #915

Merged
reardencode merged 4 commits into
masterfrom
test/shorter-pads
Oct 5, 2026
Merged

reardencode merged 4 commits into
masterfrom
test/shorter-pads

Conversation

@rearden-grok

@rearden-grok rearden-grok Bot commented Oct 5, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Index progress takes the caller's interval. Production still passes the 10s cadence. The journey holds one second and pins that constant, so a hardcoded 10s comparison or a changed literal fails.
  • The resume walk scores 256 headers on a 32 KiB thread. The production walk is a heap stack. The old 12,000-header fork only asserted a 32-header cap.
  • SV2 listener checks share one stale hub. The write-deadline pin is process-wide, so that chapter runs after the other accepts, and a stall of two seconds or more fails. Template and IBD chapters copy one 3-coinbase pad. The custom pow-limit chapter still builds its own chain.
  • Thirteen tx_relay chapters share one 3-coinbase pad and each open a fresh mempool. The 257-coinbase scan and the height-5 fee-history tests keep their own pads. The snapshot refresh still waits out its interval, and the chapter that connects a block runs last.

No operator-facing change.

Tests

Local debug --lib on this VM: the index journey no longer sleeps 12s; the resume journey finishes in 0.11s; sv2 is 13 tests in 6.02s against 20 tests in 23.58s on the previous suite; net is 638 tests, with the shared pad journey at 2.86s.

rearden-grok Bot added 4 commits October 4, 2026 20:30
The released-index journey slept PROGRESS_EVERY plus two seconds, so the
consensus lib waited out the operator's 10s cadence. Production still
passes that constant. The journey holds one second and pins the constant
at 10s, so a hardcoded 10s comparison or a changed literal fails.
The most-work resume journey inserted 12,000 headers to prove the walk
does not overflow, then asked for 32. The walk is a heap stack, so the
band is 256 headers scored on a 32 KiB thread: long enough that a
recursive walk of it does not fit, and still past the 32-cap.
Listener checks share one stale regtest hub. The write-deadline pin is
process-wide, so that chapter runs after the other accepts, and a stall
of two seconds or more fails. Template and IBD chapters copy one
3-coinbase pad instead of connecting a hundred blocks each. The custom
pow-limit chapter still builds its own chain.
Thirteen mempool chapters each connected a hundred-block pad. They now
share one 3-coinbase chain and open a fresh mempool directory. The
257-coinbase scan and the height-5 fee-history tests keep their own
pads. The snapshot refresh still waits out its interval, and the chapter
that connects a block runs last.
@reardencode
reardencode merged commit 04eb23e into master Oct 5, 2026
18 checks passed
@rearden-grok
rearden-grok Bot deleted the test/shorter-pads branch October 5, 2026 03:55
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