## meta-pytorch/monarch — v0.4.0…v0.4.1

_109 commits._

### Features
- **add async push_snapshot into TableStore (#3370)** (a0f7ef2)
- **Add parallel transfer support to TCP fallback backend (#3341)** (94359cb)
- **snapshot introspection: add NodePayload → relational row conversion (#3353)** (0819dd2)
- **exec_command: support running .py scripts and -m modules via compile/exec (#3362)** (dd0aaa6)
- **Add documentation page for the docs website (#3340)** (0ee2e38)
- **add relational Arrow schema for mesh introspection snapshots (#3338)** (3b0cc78)
- **Add Monarch butterfly favicon to browser tab (#3336)** (f91708f)
- **Add retry logic to mesh admin telemetry integration tests for opt mode reliability (#3321)** (e0389d3)
- **Add serve/exec CLI with rank targeting, streaming output, and mount recovery (#3158)** (9985228)
- **Re-add deprecated is_rdma_available function (#3298)** (f242317)
- **doc: add example of using kueue for dpp K8s (#2792)** (498b53f)
- **Add Beta badge to header** (293d1a7)
- **Add automatic GPU profiling to IR timing system (#2574)** (2eb9294)
- **Add export_dag_json_timed() for timing-accurate trace visualization (#2400)** (36a2ed1)
- **Add import_timing() to load external timing data into Simulator IR (#2399)** (ae8cff7)
- **Add gather_mount: read-only FUSE mount of remote host filesystems (#3288)** (9c9972f)
- **introduce attachment lifecycle for HostAgent (#3285)** (bea59fc)
- **add LinkStatus to send loop error messages (#3279)** (b2095e3)
- **add process lifecycle logging to ProcessJob (#3278)** (8129247)
- **Add version switcher to docs with multi-version deployment (#3236)** (79c0867)
- **Add /v1/query and /v1/pyspy_dump to mesh_admin_skill.md (#3240)** (3864bc8)
- **Add POST /v1/query and POST /v1/pyspy_dump routes (#3201)** (c1757c6)
- **Add pyspy dump routing and dashboard query/storage endpoints (#3198)** (1a3cc1f)
- **add FI-7/FI-8/SV-1 contract tests for propagated-stopped-child root-cause attribution (#3233)** (f14caeb)

### Fixes
- **exec bug fixes (#3375)** (9a836ff)
- **remotemount: fix FUSE refresh race and eliminate full chunk copy (#3364)** (1dab956)
- **Fix multi-GPU CUDA context in TCP fallback read/write (#3363)** (b4f96eb)
- **Small fix so kill works the same as apply (#3360)** (2f67275)
- **Fix stop_actor_by_name returning immediately without waiting for actor termination (#3323)** (7d6981d)
- **Fix FUSE non-empty mount and absolute path resolution for remote_mount (#3354)** (2d61315)
- **Fix mesh_controller to send accumulated responses for OncePort (#3352)** (fb2348c)
- **Fix CUDA 13 build errors (#3343)** (728a937)
- **CLI bug fixes (#3347)** (cc59141)
- **Fix proxy liveness test hang by using actor instance instead of context() (#3322)** (8bd5621)
- **Fix bugs in doc_build deploy (#3319)** (679faa7)
- **Fix broken fbcode//monarch/python/monarch/_src/actor:actor-type-checking - main caused by D98248748 (#3301)** (b0d6682)
- **Fix mlx5dv segment registration mr_addr (#3293)** (d9bde77)
- **Improve ActorSupervisionEvent display, fix double fault hook invocation, and simplify error output (#3068)** (5b91aa2)
- **Fix shape_json format inconsistency and add JSON validation test (#3273)** (6d8e20e)
- **Fix agentDisplayName null/undefined crash (#3262)** (755959e)
- **fix benchmarks (#3235)** (fee3340)
- **Fix zero-copy regression in pickle send path (#3234)** (2e688be)

### Backend
- **bundle export/import: arrow IPC bundles with BN-* invariants and round-trip tests** (d42b4a5)
- **snapshot service: capture pipeline with multi-sink API and drain_to_batches refactor** (c5a6219)
- **Enable rand_core feature for digest 0.10** (c6d9440)
- **restore comments** (d805d43)
- **Cleanup hosts/job on exec (#3365)** (42ee5df)
- **scope host mesh stop() to only drain its own procs (#3373)** (8b1d4bd)
- **clean shutdown for monarch client actor (#3368)** (61f0ef3)
- **integration test: live mesh capture -> push -> SQL (#3374)** (1292b13)
- **make TableStore::ingest_batch async to avoid leaking block_in_place into downstream Rust callers (#3369)** (214fb14)
- **Update from rand 0.8 to 0.9** (090ffa4)
- **opaque TableStore handle with refined boundary — remove table_data(), privatize ingest_batch, return Arc<dyn TableProvider> (#3359)** (a32f8c5)
- **expose DatabaseScanner::ingest_batch as public ingestion API (#3358)** (1299e22)
- **snapshot introspection: BFS capture with typed fold seam and visited-set dedup (#3357)** (328b0bb)
- **integration test proving NI-2/NI-3: actor supervision children coexist with proc membership (#3355)** (dd36dc6)
- **Bump datafusion 52.1.0 -> 52.4.0** (73243d6)
- **Bump arc-swap 1.8.2 -> 1.9.0** (27e6da1)
- **Bump opentelemetry-otlp 0.31.0 -> 0.31.1** (8ee7b47)
- **If you turn off all mounts, make sure the mount engine stops (#3361)** (5ad5041)
- **lints (#3356)** (4d31835)
- **Handle spawn failures in _spawn_missing_children (#3346)** (9adc441)
- **mesh_admin: separate HTTP DTOs from typed introspection domain types (#3342)** (741ff9d)
- **Restore debug subcommand accidentally removed by D96169395 (#3351)** (9b67c60)
- **Refactor TCP fallback test infrastructure (#3348)** (6df6fe5)
- **Deprecate ProcMesh.allocate_nonblocking in monarch test files (#3335)** (215d644)
- **Store py-spy dumps on the root telemetry actor (#3345)** (0ea0c07)
- **Remove ProcMesh._from_alloc (#3320)** (0ce98a7)
- **Controller label, tooltip improvements, and cosmetic fixes (#3315)** (5b3b8c0)
- **CLI/Demo Omnibus (#3337)** (3c5350b)
- **mesh_admin: replace stringly introspection refs internally with typed ids/time, keep HTTP API curl-friendly (#3334)** (347350d)
- **extract RecordBatchBuffer trait into shared monarch_record_batch crate (#3304)** (c2f9a4a)
- **mesh admin: spawn mesh admin on the calling actor's proc (#3290)** (769bf48)
- **Remove MockMast and MockMastAllocator (#3317)** (ed65951)
- **remotemount.refresh: live mount updates without unmounting** (8b36030)
- **Use scoped_state in more tests that use ProcessJob (#3327)** (aec5102)
- **Remove more files from docker images (#3305)** (68f6ec4)
- **Migrate more from _spawn_admin to Job integration (#3314)** (de81191)
- **Remove deprecated _stop_instance method (#3303)** (5d45847)
- **Consistent node rendering and tooltip improvements (#3313)** (006b145)
- **.scoped_state() in Job (#3000)** (4cda0bb)
- **DAG view improvements: admin API integration, collapse/expand, system filtering (#3302)** (8694dc3)
- **Export MONARCH_ADMIN_URL from _spawn_admin (#3297)** (077fe88)
- **Don't generating plots for hyperactor_mesh benchmarks (#3289)** (a7d54e3)
- **Refactor gather_fuse into a generic read-only FUSE filesystem component (#3296)** (58b5b30)
- **Integrate mesh admin with Job API (#3292)** (bcdce7b)
- **Integrate telemetry/dashboard with Job API (#3291)** (6c0b176)
- **Bump clap 4.5.60 -> 4.6.0** (d9e9245)
- **Integrate communication modeling into IR timing (#3121)** (1a8c109)
- **export_command_types (#2398)** (4d7972b)
- **clamp proc status to host agent state (#3284)** (e96c425)
- **flush pending actor spawns before ProcMesh shutdown (#3283)** (1d8ccce)
- **replace panics with errors in host agent handlers (#3282)** (83f07c2)
- **flush pending spawns before HostMesh shutdown (#3281)** (3a0ea75)
- **propagate close reason through TxStatus::Closed (#3280)** (b15f831)
- **non-blocking TerminateChildren via child actor (#3277)** (540c2a5)
- **two-phase shutdown for HostMesh (#3276)** (237eecd)
- **Bump tracing-subscriber 0.3.22 -> 0.3.23** (4223834)
- **Populate endpoint name in messages table (#3263)** (79d04f4)
- **Update ratatui from 0.29.0 to 0.30.0** (9875625)
- **Update smol_str from 0.2.2 to 0.3.6** (9e9fc6d)
- **Update smol_str from 0.1.24 to 0.2.2** (2ec08da)
- **Return (QueryEngine, telemetry_url) from start_telemetry (#3200)** (870d060)
- **monarch] [simulator] Refactor convert_msg to use data-driven field mapping (#2405)** (3295d2d)
- **Bump once_cell 1.21.3 -> 1.21.4** (63b346b)
- **ir.py refactoring (#2416)** (48d7531)
- **simulator.py refactoring (#2417)** (a892658)
- **generalize spawn_admin over a set of host meshes for multi-mesh admin aggregation (#3226)** (a9d5ed0)
- **Overwrite display name on proc and host agent (#3232)** (f8e9a76)

_Recap by [Repo Wrapped](https://repowrapped.com/gh/meta-pytorch/monarch?utm_source=github-action)._