floatdrop / zoxy / 28699233514
96%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 26
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 96.513% (+0.007%) from 96.506%
28699233514

push

github

floatdrop 
restart: hot restart — SCM_RIGHTS listener handoff, then the old process drains (Phase 4 slice 2)

Config 'handoff' names a unix socket. Old side: a dedicated thread serves
one successor — header + every worker's listener fd in one SCM_RIGHTS
cmsg — then SIGTERMs itself, funneling into the slice-1 drain; once
draining it accepts-and-drops so a late successor binds fresh. New side
adopts at startup: header validated, every fd getsockname-matched against
the configured listen address and SO_ACCEPTCONN-checked, or the whole
batch is refused and closed. Duplicated fds keep the accept queues alive
past the old workers' close — the drain-only RST window is gone (tested:
a queued connect is accepted through the duplicate after the original
closes). main now owns all listeners (bind failures fail startup).

Admin listener gained SO_REUSEPORT: the successor used to bind admin
after adopting, die on the predecessor's still-held port, and take both
processes' queues down — post-adoption startup failures are the
dangerous class, so everything fallible precedes the adopt except the
handoff rebind and admin, both overlap-safe now.

Verified: 6 new tests (fd round-trip, queue survival, wrong-address
refusal, first-boot fallback, unix-socket path, config); process-level
A→B restart under a request hammer: 928/928, zero failures.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

6560 of 6797 relevant lines covered (96.51%)

0.97 hits per line

Coverage Regressions

Lines Coverage ∆ File
3
95.15
 -0.89% obs/admin.zig
1
99.64
 0.02% config.zig
Jobs
ID Job ID Ran Files Coverage
1 28699233514.1 26
96.51
 GitHub Action Run
Source Files on build 28699233514