zlay
atproto relay in zig zlay.waow.tech
relay zig atproto
Star
Fork
Watch

Commits on main
docs: name the repo manifest as zlay's spindle binding
committed
docs: record the lalinsky zio ship in its handoff
committed
docs: describe lalinsky zio as the production backend
committed
ingest_ack: port the canceled-reader test to lalinsky zio's Runtime
committed
shutdown, broadcaster: never close a socket another fiber is parked on
committed
backend: swap zzstoatzz.io/zio for lalinsky/zio main (zig 0.17)
committed
docs: handoff for taking lalinsky/zio to prod
committed
build: adopt zat 0.6.0 and websocket 0.2.0; pin CI zig and build the production configuration
committed
api: replace favicon with relay fan-out mark on the zat.dev tile
committed
fix: keep collection pagination offsets in caller-owned storage
committed
fix: refresh runtime trust and reject compromised TLS roots
committed
docs: record rebase normalization and verified lightrail recovery
committed
fix: normalize missing rebase flag on forwarded commits
committed
docs: close final deployment observation
committed
docs: record verified commit CID deployment and migration limits
committed
xrpc: bound both sides of repository pagination
committed
sync: persist signed commit CIDs separately from MST roots
committed
docs: hand off corrected deployment observation
committed
xrpc: omit repositories without a complete commit from listRepos
committed
xrpc: allocate full listRepos responses and preserve pagination
committed
docs: record deployed zlay and interrupted probe checkpoint
committed
docs: handoff for deploying 205cf7f and moving lightrail back
committed
docs: replay handoff and backoff in design.md; zig 0.17 on the build server
committed
deps: zat v0.6.0-alpha.5 for the TLS validation-time refresh
committed
broadcaster: sleep instead of spin, replay at the consumer's pace, read playback outside the persist mutex
committed
xrpc: render listReposByCollection into an allocating writer
committed
Acknowledge processed upstream frames and drain producers before shutdown
committed
Keep queued failure signals alive through subscriber teardown
committed
Make event buffering and log rotation failure-atomic
committed
merge zig-0.17: port to the 0.17.0 release
committed
docs: merge CLAUDE.md into AGENTS.md
committed
zig 0.17: port to the 0.17.0 release
committed
docs: replay window is 2h by policy; deep replay belongs to stream
committed
gc: sweep orphaned segments; cover gc with a real-postgres test
committed
docs: pinger env vars in README, answered-ping semantic, zio liveness in thread table
committed
pinger: a ping is answered iff activity arrived after it was sent
committed
adopt websocket v0.1.13 + zat v0.3.31: close/write race fixed; stagger pings
committed
subscriber: keepalive pinger detects half-open upstream connections under zio
committed
metrics: zio block no longer silently dropped when it outgrows its buffer
committed
liveness: /_healthz 503s on a process-wide ingest stall, runtime-settable
committed
metrics: export zio_sched_inject_drops_total
committed
adopt zio a58da60: events_rearmed actually exported, inject drops counted
committed
metrics: events_rearmed help text no longer attributes the 2026-08-18 incident
committed
scrub sweep references from comments, drop stray blank metrics line
committed
Revert the silent-worker sweep: silence is the normal case, not a fault
committed
slurper: close the sweep's pre-handshake blind spot
committed
slurper: sweep workers that are connected but delivering nothing
committed
admin: guard forced reconnects behind an idle window, add dry run
committed
admin: force a host's worker to reconnect, in band
committed
metrics: actually emit zio_sched_events_rearmed_total
committed
adopt zio e1fc874: lost poller wakes are re-armed, and export the counter fix/zio-deafness-rearm
committed
validator: correct the incident comment to match the revised analysis
committed
authority checks: bound the network phase, unconflate the metrics
committed
sweep: default to the deactivated cohort, not every non-active status
committed
sweep: tolerate per-candidate network errors instead of dying
committed
fix: recover accounts muted by stale upstream account status
committed
docs: handoff for the consumer-flap OOM fix
committed
fix: bound disk replay per chunk and abort for dead consumers
committed
docs: CLAUDE.md is instructions, not project state
committed
docs: drop the retracted stack-aliasing claim
committed