KarpelesLab / libwallet / 27532682092

24%

push

github

wltwallet: retry reshare init on a different wdrone peer + bump 0.4.60

Followup to the "context deadline exceeded" reshare failure that
persists on 0.4.59 (the protocol-on-upload regression fix doesn't
cover the case where the peer wdrone picked by selectPeer is itself
stuck on /init).

A wdrone peer can pass /ping in sub-second and still hang on /init —
slow loadShare, internal goroutine wedged, spot routing wobble. Without
retry, one bad pick produces a 90-s reshare failure for the user even
when other wdrones in the fleet would have served the init promptly.

spotPeer.Start now makes up to 3 attempts, re-running selectPeer with
the previously-tried peers excluded so a retry never re-rolls into the
same bad peer. The 90 s per-attempt budget stays — wdrone's loadShare
ceiling is 60 s and the spare 30 s is response routing; bumping the
per-attempt window further would just extend the failure tail without
helping the common case. The surfaced error names every peer that was
tried so the next field report can be matched to specific wdrones.

selectPeer gains a variadic `excludes` parameter (zero excludes is
identical to the original behaviour, keeping the test caller intact)
and refuses to roll into a peer the caller already burned.

reshare_scenarios_test now widens its skip predicate to also accept
"failed to init remote: context deadline exceeded" because the
backend reachability flake on real-network test runs is dominated by
that pattern with the protocol-on-upload fix in place. The
"no payload available" substring still fails loud — that's the genuine
protocol-on-upload regression signature.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

20 of 38 new or added lines in 2 files covered. (52.63%)

4393 of 18726 relevant lines covered (23.46%)

101.94 hits per line