n1m7j6b2v07tsk28kjss631tkjs85v7b · 8 rows
| id | content_hash | title | description | design | acceptance_criteria | notes | status | priority | issue_type | assignee | estimated_minutes | created_at | created_by | owner | updated_at | closed_at | closed_by_session | external_ref | spec_id | compaction_level | compacted_at | compacted_at_commit | original_size | sender | ephemeral | wisp_type | pinned | is_template | mol_type | work_type | source_system | metadata | source_repo | close_reason | event_kind | actor | target | payload | await_type | await_id | timeout_ns | waiters | hook_bead | role_bead | agent_state | last_activity | role_type | rig | due_at | defer_until | no_history | started_at |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| tarantool-etcd-234 | 6a864d040222a5c7b7ec4e61daed84bfd693cc407841df0f2101755949bfd026 | LeaseGrantRevoke perf regression — revalidate before chasing | LeaseGrantRevoke regressed from 2 243 ops/s to 1 223 ops/s against the previous bench baseline (docs/BENCH.md "Refresh — 2026-05-14, post select+count Range fast path", section "Known issues — open"). Bench code on this leaf is unchanged; revalidate with a second run before chasing the cause — bench noise on this leaf is plausible. | 1. Re-run BenchmarkLeaseGrantRevoke on the same hardware (Apple M4 Pro per BENCH.md), 5 s/leaf, fresh subprocess per leaf. 2. If still ~1 223 ops/s, bisect across the commits between the 2 243 and 1 223 measurements (git log between BENCH.md baselines). 3. If back at ~2 243, close as noise. | Source: docs/BENCH.md section "Known issues — open" near line 244. | open | 3 | bug | NULL | NULL | 2026-05-19T13:34:48Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:34:48Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL | ||||||||||||||||||||||
| tarantool-etcd-9z2 | 7993430a7610e43d13141d366e3b7b3f7ec6ea2063030b9d2815a2a770cf3546 | Phase 5f — Watch option coverage | Per-option density on Watch RPCs. Big stream-shaped surface, highest remaining value but also highest effort. Run after Phase 5e. Likely tests: - TestWatchWithFilterPut / TestWatchWithFilterDelete (server-side event filtering) - TestWatchProgressNotifyExplicit (RequestProgress on demand vs the implicit 10-minute heartbeat already covered) - TestWatchWithFragment (large-response fragmentation — may surface a tarantool-side gap) - TestWatchWithPrevKV (per-watcher prev-kv flag, distinct from WithPrevKV on Put/Delete) - TestWatchFromCompactedRevision (ErrCompacted returned on the stream, not as a typed gRPC status) | Each listed test ported, runs on all four cells, passes (or documented as a known gap with cross-reference). | Source: docs/TODO.md "Open / Conformance matrix — Phase 5 option coverage". | open | 2 | task | NULL | NULL | 2026-05-19T13:34:06Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:34:06Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL | ||||||||||||||||||||||
| tarantool-etcd-b08 | e3928051dd644ec2d3c70348320e479a23d4e425c4f5776f5a2cbbfc3d571607 | Phase 5g — Auth/Cluster/Maintenance/Lock/Election option coverage | Per-option density on the remaining v3 services. Low priority — each surface is small and the existing files already cover the canonical paths; spot-fill only what option coverage actually reveals. Run after Phase 5f. | Spot-fill tests added where an unaddressed option diverges between cells; documented as a known gap otherwise. | Source: docs/TODO.md "Open / Conformance matrix — Phase 5 option coverage". | open | 3 | task | NULL | NULL | 2026-05-19T13:34:10Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:34:10Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL | ||||||||||||||||||||||
| tarantool-etcd-buq | 9370d11c13a9d704f0b6e3f0241d2513428257a7e3278aacfed07ddb5818c965 | Drop stale TODO-item-14 comment in app/etcd/json.lua | app/etcd/json.lua:621 has a stale comment: "/v3/election/observe — gated on TODO item 14". TODO item 14 closed in docs/TODO.md (the Election Observe broadcast is implemented; STREAM_HANDLERS at line 427 includes /v3/election/observe). The comment is now misleading; rewrite it to describe what the fallback branch actually catches (a streaming PATH without a handler registered) and remove the historical reference. | Comment at app/etcd/json.lua:621 no longer references TODO item 14 or implies /v3/election/observe is unimplemented. | Backlog. Found while auditing TODOs for the bd port on 2026-05-19. | open | 4 | task | NULL | NULL | 2026-05-19T13:34:52Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:34:52Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL | ||||||||||||||||||||||
| tarantool-etcd-j1c | d72af58dd67917cd0d6f45c09ebc8410746a5a5fa3ce0f2fc547a0845aa4d6da | /debug/pprof/heap — pprof-shape converter on top of /debug/memprof | /debug/memprof ships the raw Tarantool memprof binary trace. Open work: a pprof-shape converter — pure-Lua parser for the memprof binary, aggregate events by allocation site, emit a four-dimension perftools.profiles.Profile (alloc_objects/count, alloc_space/bytes, inuse_objects/count, inuse_space/bytes — Go heap profile shape). Implementation sketch: 1. Pure-Lua parser for the memprof binary format (header magic + tagged event stream: alloc / realloc / free / symtab). Reference is tools/memprof.lua + tools/utils/binreader.lua in upstream tarantool/tarantool (not bundled with the homebrew binary). 2. Aggregate into (stack_key → {alloc_objs, alloc_bytes, inuse_objs, inuse_bytes}) with inuse_* from correlating alloc events against later free events inside the same window. Stack keys are the (chunkname, line) chain from symtab events. 3. Encode via lib/protobuf/encoder.lua + lib/pprof/defs.lua — schema already supports multi-sample-type profiles, declare four sample_type entries. 4. New endpoint /debug/pprof/heap?seconds=N on the JSON gateway. Same shape as /debug/pprof/profile (600 s cap, 503 on memprof-unavailable). Existing /debug/memprof stays for raw-trace use cases. | Known frictions: - misc.memprof.available == false on Apple Silicon (same arm64 constraint as the JIT — see reference_apple_silicon_no_jit). start() returns true but stop() says "profiler is disabled" and writes nothing. /debug/memprof already detects this and 503s; pprof-heap should do the same. Linux/x86_64 unaffected. - inuse_* is "alive at end-of-window", not "alive right now" — long-lived objects allocated before the window started never appear. Document this; Go heap profile has the same caveat. - Writing the binary parser without ground-truth output is the main risk — develop on a Linux box where memprof actually fires, validate against tools/memprof.lua textual report. | 1. /debug/pprof/heap?seconds=N returns valid pprof-shaped protobuf, parseable by `go tool pprof`. 2. On Apple Silicon, returns 503 with clear "memprof unavailable on arm64" message. 3. Round-trip tests in test/pprof_test.lua (same shape as the cpu profile tests). | Source: docs/TODO.md "Open / Pprof-shape /debug/pprof/heap". Referenced from app/etcd/debug.lua:16, lib/pprof/memprof.lua:12. | open | 3 | feature | NULL | NULL | 2026-05-19T13:34:30Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:34:30Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL | |||||||||||||||||||||
| tarantool-etcd-lac | 9b0811dbe3090da92e2f2c059b263cbef43e1721b4418917747dcf3102a20f60 | Phase 5e — Lease option coverage | Per-option density on Lease RPCs. Canonical paths are already covered; what is missing is option coverage in the {tarantool, etcd} × {grpc, json} matrix. Smallest remaining surface — should be the warm-up before Phase 5f. Likely tests: - TestLeaseGrantWithExplicitID - TestLeaseGrantTTLBoundaries (TTL=0 / negative / very-large) - TestLeaseTimeToLiveAttachedKeys - TestLeaseRevokeRevokesAttachedKeys (each attached key disappears + watcher sees DELETE at the revoke revision) - TestLeaseKeepAliveOnceTTLZero (cross-wire — jsonclient-internal case already pinned) Sub-phases 5a-c each surfaced one Lua-side divergence — expect at least one more here. | Each listed test ported, runs on all four cells, passes (or is documented as a known gap in tests/conformance/harness/gaps.go cross-referenced from docs/TODO.md). | Source: docs/TODO.md "Open / Conformance matrix — Phase 5 option coverage". | open | 2 | task | NULL | NULL | 2026-05-19T13:34:01Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:34:01Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL | ||||||||||||||||||||||
| tarantool-etcd-lko | a9f5b3fdb1d8c955ea9817c5961cd2fd334733f9495cfcea8dc728d5d59c5bad | Route writes to current master (write-forwarding on followers) | Followers currently reject writes with ER_READONLY → gRPC UNAVAILABLE, and clientv3 does NOT fail over endpoints on Unavailable (see memory feedback_clientv3_error_assertions). Clients must pin to the leader, which breaks endpoint-agnostic client setups (AutoSyncInterval, DNS round-robin, k8s Service). Implement write-forwarding inside the role: when a follower receives a mutating RPC (Put / DeleteRange / Txn-with-writes / LeaseGrant / LeaseRevoke / LeaseKeepAlive / Compact / Auth* / Lock / Campaign / Resign / Proclaim), proxy it to the current Raft leader via experimental.connpool + a setuid stored Lua function (same pattern as MaintenanceMoveLeader, see project_cross_node_promote memory). Reads stay local per replicaset.read_pref (already implemented). | - Reuse the connpool helper introduced for MaintenanceMoveLeader (tests/conformance/harness/tarantool_rs_server.go:475 area). - Trigger forwarding inside grpc.lua make_handler before dispatching the write — cheaper than wrapping every kv/lease/txn module. - box.info.election.leader gives the leader id; map id→advertise URL via advertise.lua (already used for MemberList clientURLs). - Streaming RPCs (Watch, LeaseKeepAlive, Election Observe) are subtler — defer or handle separately; first cut should be unary-only. | 1. Mutating RPC on a follower in tarantool_rs3 backend succeeds (matches the leader-equivalent response shape, including header.revision monotonicity). 2. No regression in existing rs3 conformance subtests. 3. New conformance subtest TestWriteForwardingOnFollower covers Put/Txn/LeaseGrant from a follower endpoint. 4. If the leader is unknown (election in progress), return ErrGRPCNoLeader as today. | Discussed in chat 2026-05-19 as the only multi-master option that preserves etcd contract. Memory refs: project_cross_node_promote, feedback_clientv3_error_assertions. | open | 2 | feature | NULL | NULL | 2026-05-19T13:33:55Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:33:55Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL | |||||||||||||||||||||
| tarantool-etcd-rzr | 3973dcbaf27d8ffaf25bdad2afcfcbfeb9e6deb07cd8ecf62c5a6446247c3468 | nghttp2 unary send-path rewrite — close parallel KVGet gap, drop jit.off | Parallel KVGet on tarantool is ~70% slower than etcd at GOMAXPROCS=6 (12 688 ops/s vs 21 900 ops/s; p95 970 µs vs 405 µs — docs/BENCH.md "Linux x86_64 — KVGet matrix, 2026-05-15"). Root cause: per-Connection IO mutex in lib/nghttp2/grpc.lua (Connection:_locked) — nghttp2 sessions are not reentrant, so every fiber's submit_response + send_all + sock:write serializes. Etcd has no such constraint. Fix: pre-build HEADERS + DATA + trailers in pure Lua, skip nghttp2's C→Lua data-source callback on the unary send path, recover ~10–20% throughput lost to jit.off + drop the JIT pin entirely. | See memory project_nghttp2_send_callback_rewrite for the plan. Pre-frame DATA in Lua, avoid re-entering nghttp2 mid-send. Touch points: - lib/nghttp2/grpc.lua (Connection, _locked, send_response) - lib/nghttp2/ffi.lua (data-source callback signature) - M.recv/M.send/M.send_all/M.send_data currently pinned jit.off — un-pin once the callback re-entry is gone (memory feedback_nghttp2_session_not_reentrant explains the original panic). | 1. BenchmarkKVGetParallel/point closes the gap to within ~10% of etcd at GOMAXPROCS=6. 2. p95 latency under parallel load drops below 600 µs. 3. No bad-callback panics across 10 verbose runs (the regression test from commit d0fa396). 4. jit.off removed from M.recv/M.send/M.send_all/M.send_data. | Sources: docs/BENCH.md "Linux x86_64" + memory notes project_nghttp2_send_callback_rewrite, feedback_nghttp2_session_not_reentrant, feedback_nghttp2_dual_connection. | open | 2 | task | NULL | NULL | 2026-05-19T13:34:42Z | Eugene Blikh | bigbes@gmail.com | 2026-05-19T13:34:42Z | NULL | NULL | 0 | NULL | NULL | NULL | 0 | 0 | 0 | <binary> | 0 | NULL | NULL | NULL | 0 | NULL |