Repository navigation
id: investigate: hot-world sharding — when one world outgrows one Durable Object #2059
Description
Source: the Cloudflare Workers fit investigation (on #2044) open question 3. A per-world Durable Object has a ~1,000 req/s soft ceiling and single-threaded execution. Investigate sharding/overflow strategies for a hot world (partition by topic/region, spill to native nodes, hierarchical objects) and the signals that trigger each.
Program: Drew's 2026-10-28 directive (full text: hatch-personal memory/topics/id-program-2026-10-28.md). Parent epic: #2056.
Working pattern: parallel investigations, then the named synthesis issue merges them into a full design. Route guidance: Workhorse by default; Specialist/Summit for synthesis and ramification-heavy work. A result that says "done" is not the endpoint — the orchestrator reviews it and may commission another round. Deliverables: a markdown design/investigation doc (mermaid/SVG diagrams welcome) linked from this issue, plus follow-up issues for anything buildable. Build a proof of concept where the design calls for one, but do NOT build the permanent feature on top of an unreviewed PoC.
Acceptance criteria
- A markdown investigation doc is linked from this issue covering sharding/overflow strategies for a hot world (partition by topic/region, spill to native nodes, hierarchical objects).
- The doc names the signals that trigger each strategy, against the per-world Durable Object ~1,000 req/s soft ceiling and single-threaded execution.
- Anything buildable that falls out is filed as follow-up issues linked here.
Activity
dezren39 commentedon Oct 29, 2026
Parent epic: #2056 — ID platform expansion (2026-10-28 directive). Filed as batch 2 from PREPARED-ACTIONS.md.
developing-today commentedon Oct 29, 2026
CODE-980
dezren39 commentedon Oct 29, 2026
Intake review
Related: parent epic #2056; source: Workers fit investigation #2044, open question 3; sibling investigations #2060 (key recovery) and #2061 (jurisdiction pinning).
Notes: investigation deliverable in the ID program batch; the trigger signals matter as much as the strategies — an overflow path with no observable trigger is not a design.
Risks: partitioning by topic/region changes world consistency semantics; the doc should state what consistency each strategy gives up.
Suggested approach: Workhorse investigation per program route guidance.
— Fleet monitor intake pass (protocol §5): issue verified still live this run (filed within the last hour; no merged PR or duplicate resolves it), labels normalized, meta/enriched applied.
dezren39 commentedon Oct 29, 2026
Top priority — ID program wave-1. Drew directed this be picked up in the next runs (2026-10-29): enqueued directly as triage task 3361 in repo-health-code-pilot at queue priority 20 (only breakage-class work at priority 10 outranks it). Wave-1 is the ungated investigations and designs; syntheses and PoCs stay gated on their named inputs. Investigation route: Workhorse.
8 remaining items
dezren39 commentedon Oct 30, 2026
Queued: repo-health-code-pilot task 3928 (2026-10-30), wave-2 follow-on. The 2026-10-29 triage verified this issue and recorded 'write the investigation' as the next step; this task is that write — investigation doc PR under doc/, follow-up issues for buildable pieces, close with the doc link. Ordered by Drew ('queue unrun').
dezren39 commentedon Oct 30, 2026
🤖 Agent note · picked up
doc: fix four Codex findings — windowed queue-depth signal, re-merge …
Merge pull request #2820 from developing-today/queue/3928-2059-hot-wo…
dezren39 commentedon Oct 30, 2026
Investigation landed — closing
The wave-2 investigation/design document for this issue is landed on main:
doc/2026-10-30T19-35-00Z_investigation_hot_world_sharding/2026-10-30T19-35-00Z_investigation_hot_world_sharding.mdIt formalises the wave-1 findings recorded on this issue (task 3361), re-verified against current main (
metrics.rssignal fields,SetCrossWorldWriteinactor.rs,WorldLimitsinmod.rs). It covers: the escalation ladder (rung 0 shed non-actor load → 1 topic partition → 2 region shards → 3 native spill under a journaled lease → 4 hierarchy for reads only); the §2 signal bands with sustained-window and de-escalation discipline; the §4 comparison table with each strategy's consistency cost explicit (ordering, atomicity, "is it still one world?"); and §5 on what sharding does to the journal, with multi-writer CRDT journal splitting explicitly rejected. All four Codex review findings on the PR were verified against the code, fixed in the doc, and replied to on their threads: windowed queue-depth signal (actor_queue_depth_maxis a lifetimefetch_maxhigh-water mark — a note also landed on #2275), shard re-merge as a state-import/reconciliation design rather than journal concatenation (WorldCore::restorerejects a secondCreatedheader),command_latencycorroboration (it times the whole handler, journal persistence included), and aggregate-vs-per-object storage for topic partitioning.Follow-ups filed from the investigation:
Hand-off to #2046 (doc §8): rung-1 hosted mapping (separate DOs vs segmented tables), shard re-merge design or one-way declaration, threshold defaults confirmation against #2275's detector. The signals follow-on #2275 remains open and carries the windowed-queue-depth refinement.
🤖 Agent note · done
doc/2026-10-30T19-35-00Z_investigation_hot_world_sharding/