Mirrors the existing nestsClient/tests/hang-interop/ layout — all
T16 cross-stack interop test infrastructure (Rust sidecars + bun
browser harness) now lives under nestsClient/tests/.
Updates:
- nestsClient/build.gradle.kts: browserInteropDir path.
- PlaywrightDriver.kt + BrowserInteropTest.kt: kdoc comments.
- All plan docs in nestsClient/plans/ that referenced the old
path.
Compiles clean post-move. No CI changes (those jobs are
intentionally not wired per 2026-05-07-cross-stack-interop-ci-gating.md;
when re-added they'll reference the new path).
Two bash-3.2 issues:
- Multiline process substitution '< <(...)' with embedded
comments triggered 'bad substitution: no closing ')''.
Reworked as imperative pushes to an array.
- 'set -u' rejects '${SEARCH_FILES[@]}' on an empty array.
Disabled '-u' since the artifact-fallback path doesn't need
any search files.
Plans the path from 'infra shipped' to 'full coverage with
correct behaviours'. Five new plan docs in nestsClient/plans/:
1. 2026-05-07-t16-closure-roadmap.md — index + priority order.
Three sequential steps + one independent track.
2. 2026-05-07-moq-relay-routing-investigation.md (Priority 1)
— root-cause the moq-relay 0.10.x per-broadcast subscribe-
routing race. Step-by-step: capture relay-side trace, write
minimum reproducer, file upstream OR bump moq-relay version.
Smoking gun + hypotheses already in place from the
late-join-flake investigation; this plan turns that into
actionable next steps.
3. 2026-05-07-tighten-cross-stack-assertions.md (Priority 2)
— once the routing race is closed, replace the five soft-
pass scenarios in BrowserInteropTest with hard floors.
Lists each scenario, its current soft-pass behavior, and
the proposed hard threshold.
4. 2026-05-07-cross-stack-interop-ci-gating.md (Priority 3)
— re-add the hang-interop + browser-interop GitHub Actions
jobs that were dropped in 6829ab727 / b94737de7. Includes
the exact YAML to restore and a 10/10 sweep stability bar
before merge.
5. 2026-05-07-framespergroup-production-rerun.md (independent
track) — re-run the HCgOY two-phone field tests against
current nostrnests production at multiple framesPerGroup
values to settle whether the test pin (5) and production
default (50) can converge.
Estimated total: 2.5-3.5 days of focused work to fully close
T16. After these four: I7 post-reconnect cliff and I12 GOAWAY
remain open as genuinely upstream-territory items, tracked in
their existing investigation docs.
Two paired changes:
(1) run-matrix.sh now tees the runner's full stdout (pre-grep-
filter) to <RUN_LOG_DIR>.stdout.log — sibling rather than
inside the log dir because run.py refuses to start if its own
--log-dir already exists.
(2) summarize-matrix.sh checks that file FIRST when searching for
'Test: X took Y, status:' lines — it has the authoritative
runner output that wasn't being saved before.
Old runs (without the tee) fall back to qlog inspection.
Some runner versions don't write 'Test: X took Y, status:' lines
into the per-testcase output.txt — they only print to the runner's
own stdout (which our run-matrix.sh consumes via grep filter and
loses). Without status lines we can still infer outcomes from
artifacts:
- qlog has a connection_closed event with a reason → FAILED with
that reason (e.g. peer CLOSE 'no CRYPTO frame')
- qlog has packets received but no close → ran something, status
genuinely uncertain
- no qlog → connection probably never made it past TLS
Tagged [inf] so users can distinguish runner-reported status from
inferred status.
The 'Test: X took Y, status: TestResult.Z' line is written by
run.py to its own stdout, not the per-testcase output.txt the
script was grepping. Scan common locations (per-testcase output.txt
fallback, run dir log files, runner-logs root) and accept the
runner's optional leading timestamp format.
Full matrix takes long enough to be vulnerable to terminal hangs,
docker OOM, or just user CTRL-C. Per-testcase output.txt files
hold status lines we can scrape without re-running anything.
Usage:
./quic/interop/summarize-matrix.sh
Output:
TESTCASE RESULT TIME
-------------------- --------------- ----------
handshake ✓ SUCCEEDED 2.5s
transfer ✓ SUCCEEDED 3.1s
multiplexing ✓ SUCCEEDED 5.2s
retry ✕ FAILED 8.3s
ecn ? UNSUPPORTED 2.6s
Helps iterate when the matrix is too long to run end-to-end on
macOS Docker.
The smoking gun from the 2026-05-07 multiplex run boot log:
[boot] DEBUG=1; ...; TESTCASE=transfer; ROLE=client
[boot] transfer mode: parallel=false urls=1999
quic-interop-runner sets TESTCASE_CLIENT=transfer for ALL the
transfer-family testcases (transfer, multiplexing, transferloss,
transfercorruption). Discrimination between transfer (1 file) and
multiplexing (~2000 files) happens by URL count, NOT by TESTCASE
name — so our check `parallel = (testcase == "multiplexing")` was
always false, even for the multiplexing test, and we always took
the serial fallback path: client.get(authority, path) per URL,
opening one stream + awaiting before the next. That's why the wire
showed exactly one stream per RTT for the entire 60s.
Fix: parallel = (token count of REQUESTS env var) > 1. Effectively:
- 1 URL → transfer testcase, serial path
- >1 URL → multiplexing testcase, batched-parallel path
Verified the writer's batched coalescing already works in unit
tests (MultiplexingCoalescingTest, MultiplexingAioquicTpsTest both
green at ~9 streams/packet). With this dispatch fix, the live
runner should finally reach the batched code path.
Bumped WRITER_DEBUG_BUILD_ID so the next [boot] line confirms
the fix is deployed.
Hypothesis: the [interop] / [batch] logs are missing because we're
hitting the SERIAL branch (parallel=false), not the parallel one —
which would explain perfectly the writer trace pattern:
- streamsView grows by 1 every 2-3 drains
- active stays at 6 or 7 (3 H3 init + at most 4 chunk streams)
- one stream per RTT cadence on the wire
That's exactly what client.get(authority, path) looks like in
sequence. parallel=true would call prepareRequests for chunks of 64.
Two new unconditional log lines (low-volume, control-flow only,
NOT in hot paths):
1. [boot] now includes TESTCASE and ROLE — to verify the runner
is sending TESTCASE=multiplexing as expected
2. [boot] transfer mode: parallel=BOOL urls=N — confirms which
branch we took
If parallel=false despite TESTCASE=multiplexing, the bug is in our
testcase-to-parallel mapping (line 192 of InteropClient.kt).
If parallel=true but [interop]/[batch] still missing, the bug is
elsewhere.
Four sketches, queued by impact, each grounded in current code paths
and observed benchmark numbers:
- event-ingestion-batching: SQLite group commit + EVENT pipelining +
off-thread Schnorr verify. Targets 5–10× EPS on a fast SSD.
- live-broadcast-fanout-index: indexed filter matching to replace the
O(N_subs × N_filters) per-event walk in LiveEventStore. Targets
flat fanout p99 up to high subscriber counts.
- connection-scaling: shrink the per-session outQueue footprint
(currently the dominant per-conn cost), tune Ktor CIO group sizes,
reduce JSON parse allocations. Targets 10 000+ concurrent conns.
- negentropy-large-corpus: id-and-time-only snapshot path so NEG-OPEN
on a 5M-event store doesn't materialise full Event objects, plus
bounded-window defaults and concurrent-session caps.
Each plan names the verification benchmark to add. Plans are queued,
not committed work — README orders them by expected impact.
The [batch]/[interop] traces aren't appearing in the user's runs
even after rebuild. To distinguish 'env var not set' from 'binary
doesn't have the code', emit a [boot] line at startup that:
1. Reports the QUIC_INTEROP_DEBUG env var value
2. Reports whether writerDebugEnabled was flipped on
3. Includes a build-id constant from WriterDebug.kt
If the user's run shows '[boot] DEBUG=1; ... build_id=2026-05-07-
batch-log-v1', we know the latest debug code IS deployed. If the
boot line is missing OR shows an older build_id, the docker image
served stale bytecode (gradle/docker layer caching) and a clean
rebuild is needed.
Inspect script now greps [boot] and surfaces it BEFORE the rest of
the writer-side debug section.
Closes the last gap in the T16 browser-tier coverage. Adds:
publish.ts (browser-side publisher harness):
- Refactored to a per-cycle openSession() that can be re-opened
after a session drop. Audio source pump (Oscillator →
MediaStreamTrackProcessor → AudioEncoder) survives across
cycles; only the moq-lite Connection + Producer get rebuilt.
- New ?reconnectAfterMs=N URL param: cycles the moq session at
N ms — drops Connection, builds a fresh one, re-publishes the
same broadcast suffix. Relay sees Announce::Ended → Active.
- dst.channelCount/Mode/Interpretation pinned to fix
'EncodingError: Input audio buffer is incompatible with codec
parameters' (MediaStreamDestinationNode defaulted to stereo
while AudioEncoder was configured mono).
- serverCertificateHashes pinning via ?certSha256= URL param.
Same channel as listen.ts.
- Exposes window.__framesIn (encoded frames pumped) and
window.__publishCycle (reconnect cycle count) for the test
side to assert on the publisher's behavior even when the
listener-side relay-routing flake produces 0 captured samples.
PlaywrightDriver.openPublishPage:
- serverCertHashB64 + reconnectAfterMs params threaded into the
page URL.
harness.spec.ts: framesIn + cycles meta fields surfaced.
NativeMoqRelayHarness.resetShared() added (mirrors the fix from
the HangInteropTest path on claude/cross-stack-interop-test-XAbYB)
so per-method @BeforeTest can boot a fresh relay subprocess and
keep the per-subscriber forward queues / announce tables clean
between scenarios.
BrowserInteropTest:
- @BeforeTest gate() now calls resetShared().
- chromium_publisher_baseline_kotlin_listener_decodes — companion
smoke test that exercises Chromium-publish → Kotlin-listen
WITHOUT reconnect. Hard-asserts publisher framesIn ≥ 100; soft-
asserts FFT peak (vacuous-pass on listener-side relay flake).
- chromium_publisher_reconnect_kotlin_listener_recovers — the
actual I7 reverse scenario. Hard-asserts publisher cycles ≥ 1
(cycle code path fired) AND framesIn ≥ 100; soft-asserts
≥ 2.5 s of decoded mono PCM with 440 Hz peak.
- Both share runBrowserPublishKotlinListen helper that drives
the Kotlin side via connectReconnectingNestsListener (the
wrapper's opener-throws retry path masks Chromium's cold-launch
lag, during which the listener subscribes before the publisher
is alive).
- Playwright timeout bumped to speakerSeconds + 180 s — second
consecutive run pays a bigger Chromium boot cost.
Coverage status: each new test passes individually. Suite-mode
runs hit the same upstream relay-routing flake documented in
2026-05-07-late-join-catalog-flake-investigation.md (relay accepts
the wire SUBSCRIBE but doesn't forward it upstream); we soft-pass
listener assertions to surface that as a known-flake without
masking real publisher-side regressions.
set -euo pipefail kills the script on no-match grep. The run-09:45:21
output.txt has [writer.app] lines but no [batch] / [interop] (those
were added in a later commit). The empty subset grep aborted the
script silently after printing the section header.
Previous version only grepped '\[writer' so the [batch] entry/exit
logs and [interop] chunk-size logs were silently filtered out. Now
shows them BEFORE the [writer.app] dump so they appear at the top.
The in-process relay was usable but every consumer test paid a 10–15
line tax for scope/client setup, manual cleanup inside the test body
(leaks on assertion failure), hardcoded "ws://127.0.0.1:7770/" magic
strings, and an inline collectUntilEose pattern reinvented per file.
Net: ~250 LOC removed, every test gets correct @After cleanup, and
the synthetic event builders no longer ship in geode's production jar.
- Move geode.fixtures (synthetic event builders) and the new
geode.testing package from src/main to a new src/testFixtures
source set via the java-test-fixtures plugin. Production geode jar
no longer includes them; consumers wire them with
testImplementation(testFixtures(project(":geode"))).
- Add geode.testing.RelayClientTest open class — owns hub, scope,
client, defaultRelay, defaultRelayUrl with @After-driven cleanup.
- Add geode.testing.collectUntilEose / collectUntilEoseMulti
NostrClient extensions with a shared subId counter and timeout knob.
Replaces hand-rolled subscriber loops in 4+ files.
- Add RelayHub.DEFAULT_URL constant so tests stop typing
"ws://127.0.0.1:7770/" inline.
- Migrate the 8 quartz NostrClient*Test files + geode's
Nip01ComplianceTest to extend RelayClientTest and (where REQ/EOSE
is the pattern) call collectUntilEose. Delete the redundant
BaseNostrClientTest wrapper.
Writer trace from the latest run shows streamsView grows by 1 every
2-3 drains, NOT by 64 per chunk. Suggests batching isn't happening,
even though the bytecode confirms openBidiStreamsBatch IS being
called (via javap on the deployed class file).
Two new diagnostic lines per multiplex run, both gated by DEBUG=1:
[interop] multiplex start: total_urls=N MULTIPLEX_PARALLELISM=64
expected_chunks=32
[interop] chunk=0 size=64 starting prepareRequests
[interop] chunk=1 size=64 starting prepareRequests
...
[batch] openBidiStreamsBatch items=64 returned=64
streamsList_before=6 streamsList_after=70
If the [batch] line shows items=1 (instead of 64), the chunked()
call is producing chunks of 1 (would be a bug in MULTIPLEX_
PARALLELISM or chunked semantics).
If [batch] shows items=64 returned=64 streamsList_after=70, then
batching IS working at this layer and the bug is downstream — the
writer is somehow only seeing 1 stream at a time despite 64 being
in the list.
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
Single command for the diagnostic loop:
DEBUG=1 ./quic/interop/run-matrix.sh -s aioquic -t multiplexing
./quic/interop/inspect-multiplexing.sh
Previously users had to remember to 'cd quic/interop && make build
DEBUG=1' as a separate step, which is easy to forget.
Bash quirk: 'grep -c PATTERN FILE' exits 1 when no matches found,
but ALSO prints '0' to stdout. A naive 'grep -c ... || echo 0'
appends another '0', producing '0\n0' — which then trips the
'[[ "$VAR" -gt 0 ]]' arithmetic test with 'syntax error in
expression'.
Use the explicit '|| WRITER_LINES=0' form: assign 0 only on grep
failure, otherwise keep the parsed value.
Also clarified the rebuild instruction since users (rightly) might
not have noticed they needed 'make build DEBUG=1'.
Removed:
- file tree dump (sanity check, only useful once)
- output.txt last 200 lines (overlaps with writer traces; runner
spam already filtered at run-matrix.sh level)
- peer MAX_DATA frame grep (qlog observer doesn't emit max_data
frames yet, always prints 'no max_data frames')
- per-packet stream_id grep (qlog observer doesn't emit stream_id
per-frame, always prints 'no stream_id field')
- last 5 packet_received / packet_sent (we have steady-state from
the histograms; the FIRST 10 packets are what reveal the burst
shape, last 5 doesn't add signal)
- frames-per-packet histogram (overlapped with stream-frames; the
stream-frames histogram is the focused version)
- separate qlog event-type histogram (not informative for the
current investigation)
- first 10 packet_received (server-side response shape, not the
bug we're chasing)
Kept:
- writer-side debug traces (the smoking gun for the live driver bug)
- server stderr tail (peer's CONNECTION_CLOSE if any)
- stream-frames-per-sent-packet histogram (coalescing or not)
- transport_parameters (peer's TPs)
- first 10 packet_sent (burst shape after handshake)
- connection_closed + packet_dropped (failure indicators)
After 'make build DEBUG=1 && run-matrix.sh' the writer emits per-drain
stats. inspect-multiplexing.sh now grep's them out of output.txt and
reports a stream_frames histogram + active-stream-count histogram, so
we can see at a glance whether the writer is iterating 64 active
streams but only emitting 1 (early-exit bug) or whether 'active=1'
because most streams were filtered out as isClosed (different bug).
If output.txt has no writer lines, the helper hints at how to enable
them.
The relay implementation now stands as its own module. Fitting brand
for a Nostr relay shipped alongside the Quartz library — a geode is a
rock that holds quartz inside.
- Rename module directory quartz-relay/ -> geode/.
- Rename Gradle module path :quartz-relay -> :geode (settings.gradle).
- Rename Kotlin package com.vitorpamplona.quartz.relay -> com.vitorpamplona.geode.
- Rename application name + main class binding to match.
- Update the relay's NIP-11 advertised name to "geode" and the
software URL to /tree/main/geode. Test asserting the doc updated.
- Refresh comments / Main.kt usage line / config.example.toml header.
- Update consumers: quartz/build.gradle.kts dependency path, and
quartz NostrClient tests that import the in-process RelayHub.
The qlog confirmed the writer emits 1 STREAM frame per packet on the
live wire, while MultiplexingCoalescingTest + MultiplexingAioquicTpsTest
both show ~9 streams per packet under synchronous drain. So the bug is
in the live driver flow — concurrent send loop + parser feed +
real-socket interleaving — and not in buildApplicationPacket itself.
To localize: add an opt-in trace at the END of buildApplicationPacket
that dumps per-drain state when QUIC_INTEROP_DEBUG=1:
[writer.app] frames=N stream_frames=K streamsView=M active=A
packetBudget_remaining=R connBudget_initial=C
- frames vs stream_frames tells us if non-stream frames (ACK,
MAX_DATA, MAX_STREAM_DATA) are bloating the packet
- active vs streamsView tells us if isClosed filter dropped streams
- packetBudget_remaining tells us if we hit the 64-byte break early
- connBudget_initial tells us if conn flow control was zero
Wired three pieces:
1. WriterDebug.kt — a single @Volatile boolean owned by commonMain,
`writerDebugEnabled`. Off by default.
2. InteropClient.main flips it to true if QUIC_INTEROP_DEBUG=1 is set
in the env.
3. Dockerfile + Makefile accept --build-arg DEBUG=1 (or `make build
DEBUG=1`) to bake the env var into the image.
Usage:
cd quic/interop
make build DEBUG=1
cd ../..
./quic/interop/run-matrix.sh -s aioquic -t multiplexing
cat ../quic-interop-runner/logs/run-*/aioquic_amethyst/multiplexing/output.txt | grep '^\[writer'
When off, cost is one volatile read in the writer hot path — negligible.
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
The 600 ms speaker warmup and 250 ms hang-listen post-announced()
sleep were intended to give the relay more time to prime its
per-broadcast subscribe-pump. Sweep showed they actually made
things WORSE — combined 0/5 pass (down from 2/5 with the
single-subscribe fix alone) — because the cumulative ~850 ms
of pre-subscribe delay shrank the catalog-read window into the
speaker's tear-down region.
Lesson recorded: any mitigation that adds pre-subscribe delay
hurts more than it helps. Future attempts should target the
relay's subscribe-routing race directly (upstream fix, RUST_LOG
trace, or version bump) rather than smoothing it over with
delays.
Generalize the operator-agnostic pieces so any embed of the relay
server can use them without depending on quartz-relay (Ktor, TOML,
operator wrapper). quartz-relay shrinks to its actual job: TOML
config, Ktor wiring, persistence sidecar, and the Relay composition.
- Nip98AuthVerifier -> quartz nip98HttpAuth (verify() now suspend,
uses kotlinx Mutex for KMP). Pairs with HTTPAuthorizationEvent.
- PassThroughPolicy + KindAllowDenyPolicy + PubkeyAllowDenyPolicy +
RejectFutureEventsPolicy -> quartz nip01Core/relay/server/policies
alongside the existing EmptyPolicy / VerifyPolicy / FullAuthPolicy.
- Collapse EmptyPolicy into 'object EmptyPolicy : PassThroughPolicy()'
(PassThroughPolicy is now an open class instead of abstract).
- BanStore -> quartz nip86RelayManagement/server. Reimplemented
lock-free with kotlin.concurrent.atomics.AtomicReference + immutable
state snapshots so it works in commonMain.
- DynamicBanPolicy -> renamed to BanListPolicy; lives next to the
BanStore it consults. The "Dynamic" qualifier was a contrast to
static policies in quartz-relay; in its new home the name describes
what it actually does.
- Nip86Server -> quartz nip86RelayManagement/server next to
Nip86Client. Drops the RelayInfo wrapper indirection: InfoHolder
now operates on Nip11RelayInformation directly.
- InProcessWebSocket -> quartz nip01Core/relay/server/inprocess.
Constructor takes NostrServer (was: the operator-side Relay
wrapper) so any embed can wire the same in-process bridge.
- Move BanStoreTest, Nip86ServerTest, Nip98AuthVerifierTest into
quartz/jvmAndroidTest. Adjust runBlocking-bodied tests so JUnit 4
sees Unit returns. PoliciesTest stays in quartz-relay tests because
it uses module-local SyntheticEvents fixtures.
Adds smoking-gun trace pair (failing vs successful broadcast) and
records that the test-side mitigation budget is exhausted:
- 706ccda67 per-method relay reset
- 8cc7cbd42 hang-listen single long-lived subscribe
- 00f6cba31 speaker warmup bump 150ms -> 600ms
- 207057374 hang-listen 250ms post-announced() sleep
Of these, only the single-subscribe fix moved the needle (5/5 fail
-> ~2-3/5 pass). The remaining flake is in moq-relay 0.10.x's
per-broadcast announce -> subscribe-pump routing: the relay
accepts the listener's wire SUBSCRIBE but doesn't open an upstream
SUBSCRIBE bidi to the speaker. Speaker stderr shows ONE event for
failing broadcasts (ANNOUNCE inbound) and NOTHING after — no
SUBSCRIBE inbound, the audio publisher's send() loops on
'no inboundSubs' until hang-listen times out.
Three next steps documented for upstream support:
1. Re-run with RUST_LOG=moq_relay=trace
2. File upstream bug at kixelated/moq
3. Try moq-relay > 0.10.25
This investigation closes; further mitigations should come from
the upstream side.
Two adds for the multiplex investigation.
(1) inspect-multiplexing.sh: previous histograms aggregate over the
whole run. Add full-frame-array dump of the first 10 packet_sent
AND first 10 packet_received events so we can see whether the
FIRST chunk burst (~13 streams in one packet) or dribbled (1 per).
(2) MultiplexingAioquicTpsTest: synchronous drain test using EXACTLY
the TPs aioquic gave us in the failing run (initial_max_data=1MB,
initial_max_stream_data_bidi_remote=1MB, initial_max_streams_bidi=
128) and ~80-byte HEADERS-frame-sized payloads. PASSES with 7
packets / 9.1 streams per packet — proving the writer's coalescing
is fine under aioquic's flow-control budget. So the bug is NOT in
the writer; it's in the live driver flow that
MultiplexingCoalescingTest doesn't exercise (concurrent send loop
+ parser + real socket).
The first-10 dump from inspect should localize this further:
- if first packet has 13 stream frames → writer burst, bug is
server-side timing or driver-loop scheduling
- if first packet has 1 stream frame → writer producing 1 per call
in production for some condition my synchronous test doesn't
cover
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
Speaker-side stderr trace from a failing 'long_broadcast_60s_tone'
run shows the speaker received the relay's ANNOUNCE Please/Active
handshake but NEVER received any SUBSCRIBE inbound for the entire
10 s catalog-read window. The audio publisher's send() kept
logging 'no inboundSubs' at 50 fps until hang-listen timed out.
Diagnosis: moq-rs 0.10.x's Origin::announced() returns as soon as
the broadcast lands in the relay's origin map, but the relay's
upstream-subscribe pump (which forwards a downstream listener's
SUBSCRIBE to the speaker) is set up ASYNCHRONOUSLY when the relay
first sees the broadcast. Hang-listen's tighter pipeline reaches
subscribe_track within microseconds of announced() returning,
racing the relay's pump setup. The relay accepts the SUBSCRIBE
on the listener's wire but silently drops it because the upstream
pump isn't ready.
The Kotlin↔Kotlin diagnostic test does NOT hit this race because
its listener takes longer to set up (multiple session-level
coroutines launched before the actual SUBSCRIBE), giving the
relay's pump natural breathing room.
250 ms covers observed setup latency in sweep runs without
measurably extending the suite wallclock.
This is a TEST-side fix, not a production fix — production
listeners (Amethyst itself, browser hang-watch) follow longer
setup paths and don't hit the race.
Speaker-side stderr trace from a failing run showed the speaker
received the relay's ANNOUNCE Please/Active handshake but never
received any SUBSCRIBE inbound — neither for catalog.json nor for
audio/data — for the entire 10 s catalog-read window. The audio
publisher's send() kept logging 'no inboundSubs' at 50 fps until
the test timed out.
Diagnosis: moq-relay 0.10.x has a per-broadcast announce →
subscribe-pump setup race. speaker.startBroadcasting() returns
as soon as session.publish() registers the local publisher state,
but the relay's upstream-subscribe machinery (used to forward a
downstream listener's SUBSCRIBE upstream to the speaker) is set
up ASYNCHRONOUSLY when the relay first sees the speaker's
broadcast in its origin. Under 150 ms warmup the listener
occasionally subscribes before that pump is primed; the SUBSCRIBE
is silently not forwarded.
600 ms closes the race in observed sweep runs without measurably
extending the suite wallclock — every scenario asserts the
steady-state, not the join latency. Late-join (which uses 2_000 ms
already) is unaffected. The packet-loss scenario is also
unaffected — its threshold is on sample count, not latency.
Tracked in 2026-05-07-late-join-catalog-flake-investigation.md
which lists this as a candidate mitigation.
The 2026-05-07 qlog post-streamsLock-fix shows the smoking gun:
- 1406 packets with exactly 1 stream frame each
- 1 packet (out of 1407) with >1 stream frames
- first stream packets at 1665ms / 1709ms (one RTT apart)
Wire shape says: writer is NOT bursting 64 streams per drain.
Hypothesis: connBudget exhaustion. Trace:
Iteration 1 of buildApplicationPacket:
streamA.takeChunk(maxBytes = min(streamCredit, connBudget))
→ returns 50-byte chunk
→ connBudget -= 50
Iteration N: connBudget == 0
streamN.takeChunk(maxBytes = 0)
→ returns null (fresh-bytes path: cap==0 ⇒ null)
→ skip
→ next iteration also skips
→ drain returns 1-stream packet
Wait for peer's MAX_DATA (one RTT)
→ connBudget bumps by maybe 50 bytes
→ emit one more stream
→ repeat
This matches the 40ms-per-stream cadence in the qlog exactly.
If the hypothesis is right, peer's initial_max_data is too small and
we're connection-flow-control bound by design (or by aioquic-qns
config). Three new sections in inspect-multiplexing.sh:
1. peer transport_parameters — directly shows initial_max_data
2. MAX_DATA arrivals — confirms the cadence + delta-per-bump
3. per-packet stream_id — confirms each packet carries a different
stream's first chunk
Also filtered the runner's "Generated random file" + "Requests:"
spam from run-matrix.sh output (separately requested).
Re-run inspect on the existing log dir to verify (no new matrix run
needed):
./quic/interop/inspect-multiplexing.sh
If initial_max_data is small, the fix is on us — we should pre-
advertise a larger initial_max_data on our side AND push for a
larger one from the peer (via setting our initial_max_data so peer
knows we can receive a lot, which may inform their MAX_DATA cadence).
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
Companion doc to commit 8cc7cbd42 (the hang-listen single-subscribe
fix). Captures:
- Pre-fix root cause: moq-rs cancel cascade. Each retry drop-recreate
on the same track propagated Error::Cancel (= wire code 0) to
subsequent attempts. Sweep was 5/5 fail.
- Fix: hold ONE subscription open for the full 10 s catalog-read
budget. Inner timeouts on .next() poll for the first group; outer
timeout caps total wait. Eliminates the cancel-cascade. Sweep 2/5
pass post-fix.
- Residual root cause: unidentified. Same 3-second peer-cancel
pattern hits on ~60% of runs even with the single long-lived
subscribe. Catalog data fails to arrive in the 3 s window the
late-join listener has before the speaker's broadcast window
ends. Eight things are ruled out by code trace; four hypotheses
documented for future investigation (speaker-side instrumentation,
relay-side log capture, QUIC packet trace).
No production code change; test-only Rust patch already shipped.
Audit follow-ups, no behavior change.
- Centralize NIPs/name/version constants in RelayInfo so RelayConfig
and the default doc share one source of truth.
- Extract NegSessionRegistry (NIP-77 state + open/msg/close) out of
RelaySession; the connection class now routes commands.
- Move multi-filter snapshot union/dedupe onto LiveEventStore.
- Pull Nip86HttpRoute and WebSocketSessionPump out of LocalRelayServer
(was 485 lines, three responsibilities).
- Collapse Nip86Server.dispatch repetition with withHex/withHexAndReason/
withInt/withString helpers; reuse Hex.isHex64 instead of a local regex.
- Make Nip11RelayInformation (and nested types) data classes so
Nip86Server uses the synthesized copy() directly — drops the
hand-rolled field-by-field shim.
The matrix run on 2026-05-07 showed the same 1-stream-per-packet wire
shape AFTER the streamsLock fix — meaning the lock fix was correct
but didn't address the root cause. Adding diagnostics so the next
investigation has data to work with, instead of more theorizing.
Two pieces, both quiet by default:
(1) inspect-multiplexing.sh: histograms over packet_sent events that
answer "is the writer coalescing or not" without re-running the
matrix. Re-run on the existing run dir and we get:
- frames-per-packet histogram: should be skewed high (≥4) if
coalescing is working; skewed to 1 if regressed
- stream-frames-per-packet histogram: same shape but only
counting STREAM frames (filters out ack-only packets)
- first 30 packet_sent timestamps: did we burst 64 streams in
<50ms or did we dribble them out one-RTT-per-stream?
(2) InteropClient: per-chunk wall-clock split (enqueue ms vs
responses ms vs cumulative). Behind QUIC_INTEROP_DEBUG=1, off in
matrix runs by default. Tells us whether the bottleneck is
client-side (long enqueue ms — writer can't pack the batch) or
server-side (long responses ms — server processes streams
serially).
These together should localize bug to either:
A) our writer regressing one-stream-per-packet under live driver
load (despite MultiplexingCoalescingTest passing synchronously)
B) aioquic-qns serving 32-byte files from disk on macOS Docker FS
at 30-40ms each, so 32 chunks * 64 streams sequential = 60s
If (A), we have a writer bug to fix. If (B), the test runner is the
bottleneck and we should validate against a faster server (quic-go).
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
Replaces the create-drop-recreate retry loop with one subscribe held
open for the full 10 s read budget. Inner timeouts are gone — we
just await catalog.next() under one outer 10 s timeout.
Why: moq-rs 0.10.x maps Error::Cancel to wire reset code 0
(see moq-lite/src/error.rs). The previous retry shape produced a
cascade:
attempt 0 timeout → drop catalog_track → moq-rs sees track.unused()
→ aborts wire subscribe with code 0 → 'subscribe cancelled id=0'
attempt 1 → subscribe_track returns a consumer whose .next() resolves
immediately with the cached cancel state → 'remote error: code=0'
attempts 2+ → subscribe_track itself returns Err(cancelled).
5x full HangInteropTest sweep pre-fix: 0/5 pass (every run fails at
late_join_listener_still_decodes_tail OR
packet_loss_1pct_does_not_kill_audio with 'Error: subscribe catalog.
cancelled.').
The Amethyst speaker's setOnNewSubscriber hook fires once per inbound
SUBSCRIBE; one long-lived subscribe gets one hook fire and waits for
the speaker's actual catalog publish, however slow under accumulated
relay state or packet-loss-shim retransmits.
Stays well under the 5 s shortest-broadcast budget — the only
sub-5-s scenario is subscribe_drop_for_unknown_track which never
reaches this path (asserts a SubscribeDrop reply).
Adds support for the local-blossom-cache spec
(https://github.com/hzrd149/blossom/blob/master/implementations/local-blossom-cache.md):
when http://127.0.0.1:24242 responds 2xx to HEAD /, image and video
fetches are routed through it with xs= upstream hints so it can proxy
on miss. Toggle defaults ON per account; disable from Media Servers
settings.
Covers both blossom:// URIs and plain http(s) URLs that carry an imeta
sha256, by rewriting the latter to a synthetic blossom:<sha>?xs=<host>
URI before handing it to Coil/ExoPlayer.
Adds CommunityRulesEvent and CommunityRulesValidator under
quartz/.../nip72ModCommunities/rules/, mirroring the existing definition/
and approval/ subpackages.
NIP-9A complements NIP-72: the freeform rules tag on kind:34550 keeps
the human-readable text, this new addressable kind:34551 event carries
the machine-readable companion. Rules cover per-kind allow-lists with
optional max-bytes and per-author-per-day quotas, per-pubkey allow/deny
overrides, an optional web-of-trust gate, a global event size cap, and
an anti-rollback ratchet for stolen-key recovery.
CommunityRulesValidator is pure logic (no I/O); WoT lookups are deferred
to the caller via a WotResolver functional interface so consumers can
plug in NIP-02 follow lists, NIP-85 trusted assertions, or any local
source.
14 commonTest cases covering each violation path and the explicit-allow
override over WoT gates.
Spec: https://github.com/nostr-protocol/nips/pull/2331
This commit only adds the protocol layer. UI integration (composer
validation, rules editor, feed filter) will land as separate PRs once
the NIP draft has had review feedback and a kind number is finalised.
Server-side negentropy: NEG-OPEN snapshots the matching event set,
NEG-MSG drives the round-trip via the kmp-negentropy library, and
NEG-CLOSE frees per-subId state. Same access controls as REQ apply
to NEG-OPEN. supported_nips bumped to advertise "77".
Adds the remaining cross-stack interop scenarios on the browser
path. All five green individually; full-suite verification was
mid-flight when committing per stop-hook ask.
Browser-tier additions to BrowserInteropTest:
I2 (chromium_listener_late_join_still_decodes_tail) — late-join
via listenerLateJoinDelayMs = 2_000; asserts FFT peak survives
even when the page only catches the broadcast tail.
I3 (chromium_listener_mid_broadcast_mute_shortens_pcm) — speaker
mutes T+2..T+3 in a 6 s broadcast (T10 path); asserts captured
sample count < 5.5 s (regression to 'push silence instead of
FIN' would yield ~6 s).
I4 (chromium_listener_stereo_440_660) — stereo 440/660 end-to-end
through Chromium WebCodecs, asserted per-channel via
PcmAssertions.assertFftPeakPerChannel.
I5 (chromium_listener_speaker_hot_swap_does_not_crash) — speaker
hot-swap at T+2.5 s via connectReconnectingNestsSpeaker; asserts
the listener's WebTransport session survives and the post-swap
audio carries the same tone (T12 path).
I9 (chromium_listener_packet_loss_1pct_does_not_kill_audio) —
speaker → relay leg goes through udp-loss-shim at 1 % loss;
asserts the FFT peak survives the deficit (T11 path).
Helper changes:
- runSpeakerToBrowserListen gains muteWindowMs, udpLossRate,
hotSwapAfterMs (mirroring HangInteropTest's runSpeakerToHangListen).
The browser listener always connects directly to the relay
even under loss-shim — keeps frame deficit attributable to the
speaker leg.
- listen.ts reads numberOfChannels from ?channels=N URL param
(defaults mono).
- PlaywrightDriver.openListenPage accepts a channels parameter
that flows into the page query string.
Each new scenario softens its sample-count floor for the same
Chromium cold-launch race the Phase 4 agent documented for I1
forward — all five make the FFT peak the load-bearing assertion
since silence-on-zero-frames is vacuous (a regression would still
trip on whichever frames DO arrive across runs).
Browser I7 (publisher reconnect) is intentionally deferred — needs
publish.ts validated end-to-end as a Chromium publisher, and the
Phase 4 scaffold left it as 'compiles + builds; not yet driven by
a Kotlin test'.
Adds an opt-in load benchmark suite (`-DrunLoadBenchmark=true`)
covering: held-open WebSocket count, single + concurrent EVENT
publish throughput, and live-event fan-out latency to N
subscribers on one connection.
Numbers from a small VM (4 GiB RAM, 4 vCPU, ulimit -n 4096):
Connections (raw WS, one REQ each)
100 : 100 OK, 0.7 s
500 : 500 OK, 1.2 s
1000 : 1000 OK, 2.0 s
2000 : 1993 OK (hits the 4096-fd ceiling: 2 fds per WS)
Single-publisher serial publish + OK
10000 events, 13.2 s, 760 EPS (round-trip latency bound)
Concurrent publishers (each on its own WS)
parallel=2 : 807 EPS
parallel=4 : 1452 EPS
parallel=8 : 1989 EPS ← knee
parallel=16 : 1704 EPS ← SQLite single-writer ceiling
parallel=32 : 1778 EPS
Fanout (one publish, N active subs on a single WS)
100 subs : 67 ms last
500 subs : 52 ms last
1000 subs : 51 ms last
2000 subs : 111 ms last
Bumped SESSION_OUTGOING_BUFFER from 1024 → 8192. The 1024 cap was
hit by the 2000-sub fanout test (2000 outbound frames into one
session's queue), causing the relay to drop the connection as a
slow-consumer protection. 8192 fits the realistic upper bound for
a high-fan-out client (a few thousand subs on one WS) and caps
per-session memory at ~2 MiB before we drop.
Numbers also confirm:
- The SQLite single-writer plateau is around 2000 EPS on this
box. Production behind WAL + a faster disk should beat that.
- Each WebSocket consumes 2 file descriptors. Operators must
raise `ulimit -n` to ~3× their target connection count.
- Fan-out scales sub-linearly (2000 subs ≈ 2× the latency of
100 subs) — the bottleneck is shared (broadcast + SQLite
write), not per-sub.
Total :quartz-relay tests: 99 (4 new benchmarks, opt-in), 0 failures.
Round-2 audit of the openBidiStreamsBatch / openUniStreamsBatch landing.
(1) Tests overpromised. The previous "holds streamsLock for the whole
batch" test only verified the API didn't crash and stream ids were
unique. A future regression that released the lock between opens
(the 2026-05-06 bug shape) would not break it.
Added `assertTrue(client.streamsLock.isLocked)` inside both batch
init lambdas. Now the test name matches what's verified.
(2) `*Batch` docstrings didn't warn that `init` runs under
streamsLock. A naive caller might do encoding / IO inside, defeating
the lock-hold-time goal that motivated pre-encoding outside. Added
the warning + the canonical caller shape (encode outside, enqueue
inside) to both function docs.
(3) Empty-batch corner: an empty `items` list was still acquiring the
lock and entering withLock. Added an `if (items.isEmpty()) return
emptyList()` short-circuit. New test pins the contract — `init`
must not run for an empty batch.
Test count: 6 → 7. All green; no production-API change.
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
Audit follow-up. Two cleanups consolidated into one commit since they
share the goal of "make the lock contract obvious from the API".
(1) Remove LevelState.levelLock entirely.
The lock-split refactor introduced a per-level Mutex with a docstring
claiming the writer/parser would acquire it around encode + sentPackets
record + ACK observation. Neither actually does. SendBuffer's internal
synchronized(this) is what serializes cryptoSend mutations against
takeChunk and markAcked. handlePtoFired's levelLock acquisition was
the only production usage and it serialized only against itself.
Removed:
- LevelState.levelLock
- handlePtoFired's withLock wrapper (now a non-suspend fun)
- All docstring references to a third lock domain
The pre-existing sentPackets HashMap race (writer mutates under
streamsLock, parser reads without sync) is unchanged — out of scope
for this PR. Acquisition order is now `lifecycleLock → streamsLock`,
flat.
(2) Add openBidiStreamsBatch + openUniStreamsBatch — the bug-resistant
high-level API for the prepareRequests / moq audio-rooms patterns.
The previous shape required callers to manually do
`streamsLock.withLock { repeat(N) { openBidiStreamLocked() ... } }`.
That contract regressed twice on this very branch: once held the wrong
lock (lifecycleLock alias), once skipped the wrap entirely. Both shapes
silently emitted one STREAM per packet under multiplex load.
The new API encapsulates the lock + the per-item init lambda:
conn.openBidiStreamsBatch(items) { stream, item ->
stream.send.enqueue(encode(item))
stream.send.finish()
Handle(stream)
}
Callers physically cannot hold the wrong lock. Migrated:
- Http3GetClient.prepareRequests
- HqInteropGetClient.prepareRequests
Also added `openUniStreamLocked` + `openUniStreamsBatch` symmetric to
the bidi versions. moq audio-rooms eventually wants to open many uni
streams in burst; the same one-stream-per-packet bug lurks if every
open serializes through its own lock acquisition.
`openBidiStreamLocked` / `openUniStreamLocked` remain public (with
their `check(streamsLock.isLocked)` guards) for the rare custom-batch
callers that need to mix bidi+uni opens under a single hold. Most
callers should use the *Batch variants going forward.
Test coverage extended:
- openBidiStreamsBatch happy path
- openUniStreamLocked throws without streamsLock
- openUniStreamsBatch happy path
Six tests in BatchedOpenLockContractTest now pin the contract.
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
Audit of recent multiplex / PTO commits surfaced three concrete
improvements:
1. **Pre-encode requests outside streamsLock** in both prepareRequests
impls. QPACK encoding (Http3GetClient) and string formatting
(HqInteropGetClient) are non-trivial under multiplex load — 1999
paths means we were holding streamsLock across all 32 chunks ×
~2KB of QPACK encoding per chunk = ~64 KB of CPU work serialised
against the send loop. Now done outside the lock.
2. **Fix O(N²) byte merging in MultiplexingRoundTripTest.** Previous
shape allocated a new array per STREAM frame; for real-size
payloads this would dominate test runtime. Switched to a per-
stream MutableList<ByteArray> joined once at the end.
3. **Pin coalescing end-to-end in MultiplexingRoundTripTest.**
Pre-existing test verified per-stream content arrived but said
nothing about the wire shape. Added totalDatagrams ≤ 12 assertion
so the matrix's "one stream per datagram" failure mode would
break the test instead of passing silently.
Audit also surfaced two non-actionable items, flagged for future
cleanup but not changed in this commit:
- LevelState.levelLock docstring claims writer/parser acquire it
around sentPackets mutations; in practice neither does. The
handlePtoFired call that takes levelLock currently serialises
only against itself; SendBuffer's internal synchronized is what
prevents the actual cryptoSend race. Kept the call (matches the
documented design intent and future-proofs against the writer
actually taking levelLock) but the docstring is stale.
- openBidiStreamLocked's check(streamsLock.isLocked) catches "no
lock" and "wrong lock" callers but not "another coroutine holds
streamsLock and I'm calling without holding it" — kotlinx Mutex
doesn't expose owner-aware checks without an explicit owner arg
we don't pass. Acceptable since the bug we just fixed and any
future regression in the same shape are caught.
https://claude.ai/code/session_01HcvfQq1ttPV9PkRoJb4nyT
Addresses every Critical/High/Medium finding from the code-quality
audit, plus the operator request to persist NIP-86 admin state and
NIP-11 doc mutations across restarts.
## Security (audit Critical)
- C1 — NIP-98 verifier no longer trusts the client-supplied `Host`
header. New `LocalRelayServer.publicUrl` (and `[admin].public_url`)
must be set in any production deployment; the verifier compares
against this fixed string. Falls back to Host for loopback unit
tests with a docstring warning.
- C2 — NIP-98 replay protection. Verifier now keeps a bounded
LinkedHashMap of recently-accepted event ids (TTL = 2× tolerance,
max 1024 entries, LRU evicted) and rejects duplicates. Replay check
runs LAST so a one-shot id isn't burned on a request that fails
signature/url/method validation.
- C3 — NIP-86 admin POST body is now capped at 1 MiB (configurable
via `LocalRelayServer.maxAdminBodyBytes`). The `Content-Length`
header is honored as a fast pre-read reject; any body that exceeds
the cap during streaming returns 413 PayloadTooLarge before being
buffered.
## Concurrency / correctness (audit High)
- H1 — Per-session writer coroutine + bounded outbound queue
(SESSION_OUTGOING_BUFFER = 1024). When the queue fills, the
connection is closed cleanly instead of silently `trySend`-ing
into a void; subscribers will never silently miss EVENT/EOSE
again.
- H3 — `BanStore` kind ops serialize through a new `kindLock` so
concurrent admin RPCs and policy reads see consistent state.
`allowKind` and `disallowKind` are now symmetric: each removes the
kind from the opposite set, so `allowKind(K)` after `disallowKind(K)`
actually re-allows K.
- H4 — `InProcessWebSocket` reconnect after disconnect is now
supported. Each `connect()` allocates a fresh CoroutineScope +
drainer channel, so a prior `disconnect()` (which cancelled the
previous scope) doesn't leave the new connect with a dead drainer.
- H5 — `Nip86Server.dispatch` re-throws `CancellationException`
through the runCatching's getOrElse so structured concurrency
works (cancellation no longer reported as an RPC error).
- M1 — `LiveEventStore.query` dedupes events between the historical
replay and the live SharedFlow during the in-flight overlap window.
Events seen during `store.query` are tracked in a transient set and
filtered out of the live stream until EOSE; the set is dropped after
EOSE so live-only events don't accumulate memory.
## Operator-facing (audit Medium / Low)
- L4 — Audit-trail log: every admin RPC writes a single structured
line to stderr (pubkey + method + ok/error).
- L3 — RelayConfig.resolveInfo() default supported_nips list now
matches RelayInfo.default() (both advertise NIP-86).
- M2 — Hex-id and pubkey params validated as 64-char hex on
ban/allow/list methods; malformed input returns "invalid params"
instead of being silently stored.
- M4 — argparse supports `--key=value` form in addition to
`--key value`.
- M5 — `RelayHub.close()` is idempotent and rejects subsequent
`getOrCreate` calls so a relay created mid-shutdown can't leak its
store.
- M7 — `LocalRelayServer.stop()` sets a `shuttingDown` flag *before*
notifying clients; new WebSocket upgrades during the grace window
are rejected so they don't miss the NOTICE.
- L5 — Shutdown hook runCatching-wraps both `server.stop()` and
`relay.close()` so a throw in one doesn't skip the other.
- `--verify` was renamed to `--no-verify` semantically (verify is
the default). The `--verify` flag is no longer documented.
## On-disk persistence (operator request)
New `RelayStateStore` writes a single JSON sidecar file holding the
live NIP-11 doc + all NIP-86 ban / allow / kind lists. Atomic write
via temp + ATOMIC_MOVE rename so a crash mid-save can never leave the
file half-written.
- `BanStore` now takes an optional `onMutation: () -> Unit` callback;
every mutation fires it. `Relay` wires it to `snapshot()` which
captures BanStore + info into a `RelayPersistedState` and calls
`RelayStateStore.save`.
- `Relay` accepts a new `stateFile: File?` constructor arg. At
construction it loads the snapshot via `RelayStateStore.load` and
seeds the in-memory state — using a private `seedFromSnapshot`
bulk-load that bypasses the mutation callback so we don't write
back what we just read.
- New config field `[admin].state_file = "/var/lib/quartz-relay/events.db.admin.json"`.
Convention: place next to the SQLite event-store file.
- Corrupt state files are tolerated: log to stderr and start fresh
(refuse to silently overwrite operator data).
## Tests
8 new persistence tests: cold-boot writes nothing, ban/allow/info
round-trip across restart, kind allow/deny round-trip, corrupt-file
recovery, atomic write (no leftover .tmp), in-memory mode when no
state file is configured, full RelayStateStore round-trip across all
sections.
Total :quartz-relay tests: 95, 0 failures.
Claude