[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10,"_11":12},"actionData","errors","root",{"_284":-5,"_287":288,"_289":290},"routes/shell",{"_166":167,"_283":-5,"_284":-5,"_285":286},"routes/repo",{"_159":-7,"_160":161,"_162":163,"_164":165,"_166":167,"_168":28,"_169":28,"_170":28,"_171":28,"_172":28,"_173":28,"_174":21,"_175":21,"_176":28,"_177":-7,"_178":-7,"_179":180,"_181":182,"_15":183,"_184":-5,"_185":186,"_120":187,"_122":188,"_189":190},"routes/repo.commits",{"_13":14,"_15":16,"_17":-7,"_18":19,"_20":21,"_22":23,"_24":14,"_25":26,"_27":28,"_29":28,"_30":31,"_32":33},"filter","","sources",[],"source","ref","main","page",1,"path","crates/ltx/src/codec.rs","q","qt","grep","more",false,"pending","decorations",{"_96":97,"_98":99,"_100":101,"_102":103,"_37":104,"_105":106,"_107":108,"_109":110,"_71":111,"_112":113,"_114":115,"_116":117,"_118":119},"commits",[34,35],{"_36":71,"_38":72,"_40":73,"_42":74,"_44":75,"_46":47,"_48":76,"_50":-7},{"_36":37,"_38":39,"_40":41,"_42":43,"_44":45,"_46":47,"_48":49,"_50":-7},"oid","3f22aedd1ea4d413b93e84afb1ce385f04be84f1","subject","v0.2.0","body","celld v0.2.0 changes three foundations: how cells share memory, how\ntheir state replicates, and how a fleet stays available when its\nobject store fails or slows.\n\nIsolates and memory:\n\n- Resident cells share isolates and run turn by turn on a shared\n pool. v0.1.0 held one OS thread and one isolate for each cell.\n- A resident cell costs ~471 KB against ~3.4 MB in v0.1.0, measured\n linear to 2,500 resident cells (~1.2 GB) on one node. Memory no\n longer limits the resident population; admission and RSS shedding\n set the limit.\n- A handler that awaits I/O releases its isolate between turns, so\n co-resident cells cannot stall each other.\n- celld reclaims empty cell heaps and reuses retired isolate slots,\n so a long-lived node cannot leak isolates.\n- jemalloc is the global allocator.\n\nTwo listeners:\n\n- celld now separates the data plane from the control plane. The\n public listener (--listen) serves only your Worker's routes and\n /__celld/health. A new internal listener (--internal-listen)\n carries everything else: the operator API (/state, /shutdown) and\n all node-to-node peer traffic. Keep the internal port on a private\n network.\n- A public listener on a non-loopback address requires an explicit\n internal listener. --advertise names the internal address, and\n celld diagnose probes peers at that address.\n\nReplication and compaction:\n\n- celld compacts many small level-0 replication objects into additive\n level-1 blocks (LTX v0.5.2 block format). Compaction is on by\n default and never deletes a source object.\n- A takeover of a write-heavy cell reads tens of objects instead of\n thousands. In one measured case, a cell with 4,096 durable writes\n restored from 42 objects and 0.6 MB instead of 4,116 objects and\n 19 MB, 3.8 times faster.\n- A restore downloads the objects in its plan concurrently, under one\n ceiling for the whole node.\n- A takeover pins the exact point in the replicated history that it\n restored from. celld never restores data that a fenced former owner\n uploads after that point, so a cell's history cannot fork and a\n failed write cannot reappear after a failover.\n- Compaction runs off the request path with a bounded duty cycle. A\n node that restarts with a large uncompacted backlog drains it\n without blocking durable writes.\n- celld and Litestream v0.5.16 can read both block and frame files.\n Litestream v0.5.11 can read only frame files.\n\nAvailability under storage faults:\n\n- The time that remains on the last renewed lease bounds each lease\n renewal attempt, so one hung conditional write can no longer use\n the whole authority window. In a test that silently dropped all\n established storage connections for 20 seconds, a v0.1.0-class node\n fenced; a v0.2.0 node kept serving.\n- The core event loop owns the lease timer on a dedicated thread.\n Restores, compaction, and diagnostics cannot delay it.\n- celld logs each lease attempt at start and at completion, so a\n storage hang is distinguishable from a timer failure.\n\nShutdown and restart:\n\n- SIGTERM starts a graceful drain. The drain hands each cell to a\n peer directly, so takeover does not wait for the lease TTL.\n- POST /shutdown?handoff=preserve on the internal listener prepares a\n same-node restart. The replacement process resumes its local cell\n databases and performs no owner or replica reads.\n- The drain waits for its ownership release writes and bounds each\n connection, so one idle connection cannot use the full deadline.\n- The process exits without unwinding. This fixes a crash when a warm\n node stops.\n\nStorage backends:\n\n- A gs:// bucket selects Cloud Storage's generation-based\n compare-and-swap and authenticates with Application Default\n Credentials.\n- A bucket can take a key prefix, so several independent fleets can\n share one bucket without collisions.\n\nWorkers surface:\n\n- Wasm modules deploy and load as compiled WebAssembly.Module values.\n A workers-rs example shows the path, and older nodes refuse the new\n wasm-v1 deployment feature at deploy time.\n- A new RPC example exercises entrypoints, callbacks, RpcTarget\n returns, and promise pipelining end to end. Cross-isolate stubs\n remain limited to single method calls on named entrypoints.\n- Web Crypto and node:crypto expand: asymmetric import and export\n (SPKI, PKCS#8, JWK, PEM), generateKey, P-384 and P-521, Ed25519 and\n X25519, ECDH derivation, AES-CBC and AES-CTR, DigestStream,\n timingSafeEqual, and one-shot sign and verify.\n- setWebSocketAutoResponse answers ping-style messages without waking\n the cell.\n\nOpenTelemetry support (off by default):\n\n- CELLD_OTEL=1 turns on OpenTelemetry traces and logs with no\n collector required: celld writes them directly to the bucket as\n Parquet under a telemetry/ prefix, and DuckDB can query them.\n- A standard OTLP/HTTP collector is the alternative sink\n (CELLD_OTEL_SINK=otlp), and the usual OTEL_* variables apply.\n- W3C trace context propagates end to end: celld adopts an inbound\n traceparent, and fetch spans, cell dispatch, and console logs join\n the caller's trace.\n\nOverload:\n\n- A stateless request that finds every slot held now waits, bounded\n by CELLD_ADMISSION_WAIT_MS, instead of an immediate refusal. At\n overload, the node no longer spends its capacity on refusals.\n- CPU-based shedding is removed. RSS shedding (on by default since\n v0.1.0) is the only pressure trigger.\n\nConfiguration:\n\n- celld validates every variable at startup, and an invalid value\n stops startup. Boolean variables accept only 0 or 1.\n- Removed: CELLD_WORKERS, CELLD_MAX_COHOSTED, CELLD_MAX_CPU_PERCENT,\n CELLD_RESIDENT_LOW_WATER (ignored since v0.1.0), CELLD_VALIDATE,\n and the /js route.\n- Renamed: CELLD_HIBERNATIONS is now CELLD_EVICTIONS.\n- Changed: CELLD_IDLE_EVICT_S no longer defaults to 300; idle\n eviction is off unless set. CELLD_UNSAFE_PUBLIC_ADVERTISE takes 1,\n not on.\n- Added, among others: CELLD_INTERNAL_ADDR, CELLD_MAX_STATELESS_ISOLATES,\n CELLD_ADMISSION_WAIT_MS, CELLD_LTX_COMPACTION, CELLD_SHUTDOWN_DRAIN_MS,\n and the CELLD_OTEL family.\n\nUpgrading from v0.1.0:\n\n- A v0.1.0 command line with a non-loopback --listen refuses to start\n until an explicit --internal-listen is added.\n- The upgrade must not be a rolling update: stop every v0.1.0 node,\n then start the v0.2.0 nodes. v0.2.0 advertises the internal\n listener, which v0.1.0 peers cannot follow, and writes compacted\n block objects, which a v0.1.0 reader cannot restore. A fleet must\n not mix the two versions.\n\nCo-authored-by: Nathan Whitaker \nCo-authored-by: Connor Hindley \nCo-authored-by: Yusuke Tanaka \nCo-authored-by: Heyang Zhou \nCo-authored-by: Daniel Lee ","authors",[51,52,53,54,55,56],"time",1786572757,"signature","ssh","checks",[],"touched",{"_57":69,"_59":70},{"_57":67,"_59":68},{"_57":65,"_59":66},{"_57":63,"_59":64},{"_57":61,"_59":62},{"_57":58,"_59":60},"name","Daniel Lee","email","danielylee@google.com","Heyang Zhou","hello@su3.io","Yusuke Tanaka","yusuktan@maguro.dev","Connor Hindley","conn.hindley@gmail.com","Nathan Whitaker","nathan@deno.com","Ryan Dahl","ry@tinyclouds.org","10cb1303dac710dcb3b557e318e08c855261f68b","v0.4.1","You can do a rolling upgrade from v0.4.0.\n\nHighlights:\n\n- **Balance cell ownership across the fleet.** A node that joins takes\n hibernated cells from the nodes that hold the most, so it carries its share\n within minutes instead of waiting for new traffic. The fleet evens out\n again after a node leaves. One node samples the leases for the whole fleet,\n so the cost does not grow with the square of the fleet.\n- **Restore a large cell page by page.** The cell reads each page from the\n object store on first use, so it serves requests before the download\n completes. A large cell also hands off through its LTX chain instead of a\n whole-database snapshot.\n- **Queue throughput improves by an order of magnitude.** Producer calls\n share transactions and durability rounds, and each message gets a\n time-ordered id. One Queue sustained 7,357 sends per second over a\n 300,000-send soak with exact delivery, up from a few hundred.\n- **Fewer false durability timeouts under load.** The budget measures a\n stall: a queued write waits while the node keeps landing proofs. A follower\n commits a burst of frames as one fsync chain, and CRC64 verification uses\n SIMD.\n- **Alarms scale to the due work.** The waker lists one bucket per armed\n minute and opens only the due ones. Only the elected waker probes the\n fleet, and the activation bound derives from store latency, so alarms of\n evicted cells fire on time under load.\n- **Recover from a whole-fleet restart.** A restarting node serves follower\n fragments while it recovers its own predecessor, so nodes that restart\n together recover acknowledged writes. Recovery checkpoints its progress and\n holds a heartbeat, so a retry skips finished work and peers do not repeat\n it.\n- **Durable Object facets.** `ctx.facets` provides `get()`, `abort()`, and\n `delete()`. Each facet gets an isolated SQLite database that celld\n replicates with the root object.\n- **More Node.js and web platform APIs.** `node:fs`, `node:os`,\n `node:diagnostics_channel`, a read-only `/bundle`, and a per-request\n `/tmp`. `HTMLRewriter`, `EventSource`, `MessageChannel`, the `Cache` API\n surface, outbound TCP through `cloudflare:sockets`, and Web Crypto KDF\n derivation, key wrapping, and RSA signing.\n- **A tighter output gate.** An error answer, each chunk of a streaming body,\n a call into a loaded Worker, and the answer of a handler that fails after\n it commits all wait for the proof of the writes they depend on.\n `storage.sync()` returns after the cell's writes are proven durable, and a\n rollback discards writes submitted without `await`.\n- **A lease reports why it lost authority.** A self-fence names its cause,\n and a failed renewal retries before the authority expires.\n- **Operator commands page.** A cell listing and a KV bulk export no longer\n load a namespace into memory. A lease records store outcomes and owned\n cell count, so `celld diagnose` shows each node's load. `celld dev --clean`\n discards the local state.\n\nIssues fixed:\n\n- #176\n- #178\n- #179\n- #180\n- #182\n- #183\n- #184\n- #185\n- #186\n\nCo-authored-by: Bartek Iwańczuk \nCo-authored-by: Bert Belder \nCo-authored-by: Daniel Acuna \nCo-authored-by: Daniel Levi-Minzi <51272568+dleviminzi@users.noreply.github.com>\nCo-authored-by: Josh Collinsworth \nCo-authored-by: Nathan Whitaker <17734409+nathanwhit@users.noreply.github.com>\nCo-authored-by: Yusuke Tanaka ",[77,78,79,80,81,82,83,84],1788624694,[],{"_57":69,"_59":70},{"_57":94,"_59":95},{"_57":92,"_59":93},{"_57":90,"_59":91},{"_57":88,"_59":89},{"_57":86,"_59":87},{"_57":67,"_59":85},{"_57":63,"_59":64},"17734409+nathanwhit@users.noreply.github.com","Josh Collinsworth","josh@deno.com","Daniel Levi-Minzi","51272568+dleviminzi@users.noreply.github.com","Daniel Acuna","daniel.acuna@instacart.com","Bert Belder","bertbelder@gmail.com","Bartek Iwańczuk","biwanczuk@gmail.com","90b43017241f81189453d326d05948f388b34652",{"_120":156,"_122":157},"44cec221c6d40eaa2c7cf064ff4b22bb58f7cc8b",{"_120":153,"_122":154},"1d3f5bd31c91ab60159fd265984e9c8b7ed90315",{"_120":150,"_122":151},"553ae73f83c87c3f7c7a5f73c32c2211d9d7341f",{"_120":147,"_122":148},{"_120":145,"_122":146},"ae8fac053d79f971bfcb996054bb43eb2f9b05da",{"_120":142,"_122":143},"89e4ffc53a14ecb496d2ca5014ff9d19b0061ad9",{"_120":139,"_122":140},"a52f9905425bc41134d817694bdc2c50bcc5e856",{"_120":136,"_122":137},{"_120":134,"_122":135},"12d5b6333fe52717325addcfe1e99e9fd4f77bcd",{"_120":131,"_122":132},"42269c121c989c65c0638ab01f368baf18a5f0df",{"_120":128,"_122":129},"bad4649d01f0db84cdc9093527e72e64ca7a14bf",{"_120":125,"_122":126},"f2bf648663a610eefde71f3547ad61e9b896b1f0",{"_120":121,"_122":123},"branches",[],"tags",[124],"v0.6.1",[],[127],"v0.6.0",[],[130],"v0.5.1",[],[133],"v0.5.0",[],[72],[],[138],"v0.4.0",[],[141],"v0.3.0",[],[144],"v0.2.1",[],[39],[],[149],"v0.1.0",[],[152],"v0.0.2",[],[155],"v0.0.1",[19],[158],"v0.6.2","documentNavigation","repo","denoland/celld","repoId","a23b8a10-92cf-4dc4-a7ab-a04e58298dde","ownerId","org:42048915","site","nexthub","private","admin","settings","maintainer","triage","starred","stars","watchers","watching","pinned","notify","views",[239,240,241,242,243,244,245,246,247,248,249,250,251],"navigation",{"_179":180,"_198":199,"_200":201,"_202":203,"_204":205,"_206":207,"_208":209},[],"viewer","meta",{"_191":192,"_193":194,"_195":14,"_168":28,"_196":19,"_197":28,"_174":21,"_175":21},[19],[155,152,149,39,144,141,138,72,133,130,127,124,158],"cloneUrl","https://git.nexthub.sh/denoland/celld.git","description","self-hosted, distributed Durable Objects","owner","denoland","section","head","empty","pins",[222,223,224],"documents",{"_217":218,"_219":220},"code",{"_32":215,"_216":211},"automations",{"_213":21,"_214":-5},"delivery",{"_210":211,"_212":-5},"tag","7gTA52Z9a1vO6FP6eCQPix9zO4O","releases",0,"applications","workflows","deployments",13,"changes","counts",{"_221":220},"total",38,"issue",{"_225":216,"_57":237,"_22":216,"_228":238,"_177":230,"_231":28,"_232":230,"_233":211},{"_225":234,"_57":235,"_22":234,"_228":236,"_177":230,"_231":28,"_232":230,"_233":211},{"_225":226,"_57":227,"_22":226,"_228":229,"_177":230,"_231":28,"_232":230,"_233":220},"id","issues","Issues","filters",{},true,"saved","pinnable","count","tasks","Tasks",{},"Changes",{},{"_225":216,"_57":237,"_22":216,"_228":282,"_177":230,"_231":28,"_232":230},{"_225":234,"_57":235,"_22":234,"_228":281,"_177":230,"_231":28,"_232":230},{"_225":226,"_57":227,"_22":226,"_228":280,"_177":230,"_231":28,"_232":230},{"_225":277,"_57":278,"_22":277,"_228":279,"_177":28,"_231":28,"_232":230},{"_225":274,"_57":275,"_22":274,"_228":276,"_177":28,"_231":28,"_232":230},{"_225":213,"_57":272,"_22":204,"_228":273,"_177":28,"_231":28,"_232":230},{"_225":214,"_57":269,"_22":270,"_228":271,"_177":28,"_231":28,"_232":230},{"_225":210,"_57":267,"_22":210,"_228":268,"_177":28,"_231":28,"_232":230},{"_225":263,"_57":264,"_22":265,"_228":266,"_177":28,"_231":28,"_232":230},{"_225":200,"_57":261,"_22":200,"_228":262,"_177":28,"_231":28,"_232":28},{"_225":258,"_57":259,"_22":258,"_228":260,"_177":28,"_231":28,"_232":28},{"_225":255,"_57":256,"_22":255,"_228":257,"_177":28,"_231":28,"_232":28},{"_225":252,"_57":253,"_22":252,"_228":254,"_177":28,"_231":28,"_232":28},"decisions","Decisions",{},"changelog","Changelog",{},"roadmap","Roadmap",{},"Documents",{},"activity","Activity","insights",{},"Releases",{},"Deployments","automations/deployments",{},"Workflows",{},"discussions","Discussions",{},"goals","Goals",{},{},{},{},"groups","user","agents",[],"now",1794488826390,"sidebar","expanded"]

Commits on September 5, 2026

Commits on August 12, 2026