test(daemon): add fake_dragonxd, a slow dragonxd stand-in for stop/kill testing

A scriptable fake dragonxd that reproduces, in minutes, the incident where
the GUI hard-killed a node busy with an hour-long witness rebuild:

- dragonxd's HTTP-RPC queue (8 workers + 16 queued, -rpcthreads /
  -rpcworkqueue honoured): a full queue answers an immediate plain-text
  500 "Work queue depth exceeded", and a request whose client gave up
  keeps its slot until a worker reaches it
- a global cs_main held by a "witness rebuild" that logs the real
  markers; every method except stop/help takes it
- warmup -28 with the daemon's allow-list, an optional startup replay
- dragonxd's shutdown order: stop (or SIGTERM on POSIX) unbinds RPC,
  503s open connections, waits for the rebuild, then logs
  "Shutdown: In progress..." / "Shutdown: done" and exits 0

Everything is appended to <datadir>/fake_dragonxd.events; a missing
"clean-exit" line means the process was killed. One portable C++17 file
(POSIX + Winsock), OFF by default (-DDRAGONX_BUILD_FAKE_DAEMON=ON), built
into <build>/fake_dragonxd/ so it can't shadow the real bin/dragonxd.
run-gui-with-fake.sh runs the GUI against it in a throwaway HOME.

Against master's GUI on Linux it reproduces the bug: closing during a
rebuild queues stop behind cs_main-blocked workers, then the GUI sends
SIGTERM and SIGKILL mid-rebuild, with no clean exit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This commit is contained in:

DanS

2026-10-01 00:24:42 -05:00

co-authored by

Claude Opus 5.5

parent 46189da0da

commit 7b6e8a8a8e

4 changed files with 1549 additions and 0 deletions

tools/fake_dragonxd/fake_dragonxd.cpp

+1306

View File

File diff suppressed because it is too large Load Diff