Unified ingestion workflow: streaming daemon — Slice 3 (tx-hash)

[chowbao]

Slice 3 of the unified ingestion workflow (design #722 / roadmap #777), stacked on slice 2 (#803 — ledgers + events). Base is streaming-slice2-events; the diff is the tx-hash subsystem on top. This is the largest slice — the design's hardest-to-review component — and it completes the daemon: with slices 1–3 applied, the fullhistory tree is byte-identical to the full 3-type implementation.

What this adds (the tx-hash data type + rolling index)

• The txhash column family in the per-chunk hot DB — one atomic synced WriteBatch per ledger now carries ledgers + events + tx-hashes.
• The per-chunk sorted .bin run + the per-window streamhash .idx, with the rolling rebuild on each chunk boundary, coverage [lo,hi], and the atomic promote/demote commit batch.
• The resolver's per-window IndexBuild + the executor's index-build stratum (the chunk→index done-channel dependency); index-aware discard (a hot DB survives until its window index covers the chunk), prune's redundant-.bin branch, surgical recovery of index keys, and the audit INV-2 (single frozen coverage / no leftover .bin) + INV-3/INV-4 index walks.
• The chunks_per_txhash_index config pin, rebuild observability, and the multi-window tx-hash lookup E2E (cross-window false-positive rejection).

Because tx-hash is the per-window index subsystem (coverage, commit-batch atomicity, the chunk→index DAG), this slice necessarily touches more of the orchestration than slices 1–2 — which is exactly why it was sequenced last.

Composes (already on feature/full-history)

The txhash store + single-index build (#728/#729) and the read path (#794), plus the streamhash dependency.

The stack is now complete

slices 1 (ledgers) → 2 (events) → 3 (tx-hash) reconstruct the full daemon.

Testing

Built against RocksDB 10.9.1 (grocksdb 1.10.7). Full fullhistory tree green on the non-short suite incl. the multi-window lookup E2E: go test ./cmd/stellar-rpc/internal/fullhistory/....

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.

[chowbao]

Superseded by a new 2-phase stacked series that re-slices this work by phase (backfill → live ingestion + lifecycle), with the MVP scope cuts (recovery / audit / convergence / retention-reconfiguration dropped) and the folded-in fixes (cold+hot ingest service + NewPrometheusSink metrics wiring, exponential withRetries backoff, deletion of the dead RunHot/RunCold stream-drain orchestration):

• Streaming daemon: Phase 1 layer 1 — foundations (#815) #817 → Streaming daemon: Phase 1 layer 2 — primitives (#815) #818 → Streaming daemon: Phase 1 layer 3 — orchestration + daemon (closes #815) #819 (closes Streaming daemon: Phase 1 — Backfill (catch-up + all primitives) #815)
• Streaming daemon: Phase 2 layer 1 — hot store + lifecycle (#816) #820 → Streaming daemon: Phase 2 layer 2 — live ingestion (closes #816, #808) #821 (closes Streaming daemon: Phase 2 — Live ingestion + lifecycle #816, closes [Epic] Unified ingestion workflow: full-history streaming daemon #808)

Each layer builds + go vet + go test -short green; the capstone (#821) also passes the lifecycle E2E. Leaving this open for now — can be closed once the new stack is reviewed.