kepr / 9d3def6abb3f

fix: retry Postgres on startup + restore panic stack addresses (v1.56.245) - main.zig: both PG pool inits retry up to 60s instead of exiting on first failure — after host reboots kepr can beat Postgres to its socket (seen live 2026-08-06: first start after boot failed FileNotFound; only systemd's 5s restart rescued it). Same exit(2) + message after the window. - main.zig std_options: allow_stack_tracing = true. Stripped binaries printed 'stack tracing is disabled' for all three live panics since v1.56.244 (2x @memcpy arguments alias, 1x reached unreachable code); now they print return addresses. Verified empirically: stripped + forced tracing yields symbolizable addresses with no DWARF parse, so no OOM risk. - build.zig: -Dno-strip builds a DWARF twin for offline addr2line against those addresses; deployed binary stays stripped (panic-handler DWARF parse expands to ~1.8 GB and OOMs the 2 GB VPS). Gates: zig fmt, zig build test (no-PG 662), zig build, musl ReleaseSafe cross-build all green; retry loop live-tested on tylee with a bad socket (60s of retries, then original error, exit 2).

dev · 2 weeks ago · 2026-08-06 · 15.2 MB

parent: ec348dc65bde