AGENTS.md

1### Checkpoint Channel Modes (`full` / `delta`)
2 
3Checkpointer storage runs in one of two channel modes, selected by `checkpoint_channel_mode` in `config.yaml` (default `full`). `delta` mode adopts LangGraph 1.2's `DeltaChannel` for `messages`: checkpoints store a sentinel + per-step writes instead of the full message list, so storage/serde grows O(N) instead of O(N²) in turns. All checkpointer backends (memory/sqlite/postgres) serve both modes unchanged — the semantics live in the compiled graph's channel table, not in the saver.
4 
5**Mode is process-frozen and restart-required.** `make_lead_agent` and the embedded `DeerFlowClient` freeze the resolved mode (`runtime/checkpoint_mode.py::freeze_checkpoint_channel_mode`) before compiling the graph with the mode-matched schema (`agents/thread_state.py::get_thread_state_schema`, plus `adapt_state_schema_for_mode` / `normalize_middleware_state_schemas` for middleware state). Adapted middleware schemas are cached by schema, mode, and resolved snapshot frequency so a pre-freeze ephemeral graph cannot leave a stale default-frequency schema behind. A second, different mode or frequency in the same process raises `CheckpointModeReconfigurationError`. To switch: edit config, restart.
6 
7**Delta snapshot cadence is configurable but frozen with the mode.** `database.checkpoint_delta.snapshot_frequency` (default `10`) sets the `DeltaChannel` snapshot cadence. It is frozen alongside the mode (`freeze_checkpoint_snapshot_frequency`; non-positive direct inputs raise `ValueError`, while a frozen-value mismatch raises `CheckpointModeReconfigurationError`), restart-required, and must match across every process sharing one checkpoint database — the cadence lives in each compiled graph's channel table and is deliberately NOT stamped into checkpoint metadata, so the mode-compatibility marker and full -> delta migration semantics are unchanged. Schema helpers resolve it explicit-arg -> frozen -> default, and every schema/graph cache (`_delta_thread_state_schema`, `_adapt_state_schema_for_delta`, the client agent-config key, the gateway accessor-graph cache) keys on the resolved value.
8 
9**Compiled-graph cache cap is configurable and hot-reloadable.** `database.checkpoint_graph_cache.accessor_graph_max` (default `64`) bounds the gateway accessor-graph cache, which clears wholesale at the cap. The cap is re-read on every eviction check (`resolve_checkpoint_graph_cache_max`), so a config.yaml reload takes effect without a restart — a size change never affects graph semantics, only eviction timing.
10 
11**Compatibility is asymmetric and fail-closed.** Every checkpoint written in delta mode carries metadata marker `deerflow_checkpoint_channel_mode: "delta"` (injected via `inject_checkpoint_mode`; absence of marker = full, so pre-feature checkpoints need no migration). Before any state read/write, `ensure_checkpoint_mode_compatible` rejects a full-mode process opening a delta thread with `CheckpointModeMismatchError` (surfaced as HTTP 409 with the cause and thread id by the threads router; `CheckpointModeReconfigurationError` maps to 503) — a full-mode raw read of a delta blob would silently return empty/partial `messages`. The reverse direction is allowed: delta-mode processes read full checkpoints transparently (old full checkpoints seed the delta channel), so full → delta is the smooth migration path; delta → full requires materializing/converting the data first. Detection also honors upstream's `counters_since_delta_snapshot.messages` metadata, and an explicit config marker takes precedence over any ambient context value.
12 
13**Never bypass `CheckpointStateAccessor` (`runtime/checkpoint_state.py`) for thread-state access.** It is the single choke point binding graph + checkpointer + mode: it injects the mode marker into configs, runs the compatibility check before every `get`/`update`/`history`, and returns materialized state (delta checkpoints lack `channel_values.messages` — raw `get_tuple` reads see a sentinel). Gateway `services.py` builds and passes the accessor; thread-owned reads (state/history/regeneration) must use `build_thread_checkpoint_state_accessor` so the recorded assistant's middleware schema materializes every channel. `history(limit)` semantics: `0` means zero items (explicit empty), `None` means unlimited — do not pass `limit=0` through to `graph.get_state_history`. Assistant metadata lookup is fail-closed for mutation accessors so a store outage cannot silently select the default schema and discard extension channels. In `full` mode the read path degrades to a raw checkpointer read (`_RawCheckpointReadAccessor`) when the agent factory cannot build the graph (bad model config, MCP outage) — full checkpoints carry complete `channel_values`, so reads don't need the graph; degraded snapshots take `created_at` from the standard checkpoint `ts` field, falling back to metadata only for compatibility. The delta gate still applies on the degraded path; `next`/`tasks` degrade to empty and thread status falls back to the stored status because task presence is not derivable, while delta mode has no fallback (materialization needs the channel table).
14 
15**Replay checkpoint lookup prefers lineage and degrades only for an explicitly missing legacy parent link.** Branch and regenerate paths first walk `parent_config`, which prevents a global chronological scan from selecting a sibling created by regeneration. `CheckpointParentMissingError` alone enables the bounded newest-first history fallback in `app/gateway/checkpoint_lineage.py`; cycles, dangling/non-addressable parents, target mismatches, and depth exhaustion raise `CheckpointLineageIntegrityError` and fail closed instead of selecting a sibling. The compatibility scans request 400 raw checkpoints so up to 200 duration-only entries do not consume the effective branch-history budget; the fallback scans oldest-to-newest internally, skips duration-only checkpoints, and accepts only checkpoints with an addressable id as the replay base. A source history with no discoverable pre-user checkpoint preserves the historical single-checkpoint branch behavior instead of rejecting the branch; regeneration remains unavailable for that inherited response. Existing single-checkpoint branches are not mutated by regenerate preparation, and no raw checkpoint tuple is copied across threads because delta state depends on ancestry and pending writes. Regenerate source-run lookup uses the current thread's exact event, then the server-stamped `run_id` on the copied human message, then verified RunManager content matching; it does not read parent-thread events. When an interrupted response was streamed but never checkpointed, regeneration accepts only the latest visible human message's server-stamped `run_id` after verifying that it belongs to the same thread and still has `interrupted` status. Storage or checkpoint-mode failures are not treated as a missing base and still fail closed.
16 
17**A delta-mode run cannot fork; `runtime/runs/worker.py` linearizes the resume instead.** Resuming from an older checkpoint (regenerate, or any client-supplied `checkpoint`) forks the lineage, and delta state for a fork is not materializable: `BaseCheckpointSaver.get_delta_channel_history` — and the bespoke overrides in `InMemorySaver`/`PostgresSaver` — collect **every** `pending_writes` entry stored on each on-path ancestor, but a shared parent also carries the writes of the sibling child that was abandoned. Those writes replay into the fork, so the run starts from a message list still containing the answer it was supposed to replace (#4458: regenerating in a branched thread showed the superseded assistant message beside the new one after a reload; reproduced on postgres, sqlite, and the in-memory saver). Write-to-child ownership belongs to the upstream delta contract, so DeerFlow does not reimplement the walk: `_linearize_delta_checkpoint_resume` materializes the requested checkpoint's complete state and writes every channel onto the **current head** (which has no siblings) through the state mutation graph, using `Overwrite` for reducer channels and resetting newer head-only channels to their schema default (or `None` when no constructible default exists); it then drops the `checkpoint_id` selector and lets the run proceed linearly, while the abandoned turn stays in history as the rewritten head's ancestry. The worker holds `_checkpoint_thread_lock` across `_capture_rollback_point` and the optional linear rewrite, making the rollback snapshot and rewrite atomic with graph streaming and the preceding run's duration-metadata checkpoint write. Capture preserves the complete real pre-run state; cancel-with-rollback then linearly replaces the current delta head with that captured state rather than forking the now-shared pre-run checkpoint, so the abandoned turn is restored without replaying the resume sibling's writes. The worker also recomputes the current-run message boundary from the rewritten state and fails closed (an unreadable resume checkpoint raises rather than falling back to the corrupt fork). `full` mode keeps forking — its checkpoints carry complete `channel_values` and need no replay — so LangGraph branching semantics are unchanged there. Root namespace only; subgraph namespaces are left alone.
18 
19**Wholesale state replacement uses a state-only mutation graph + `Overwrite`.** `update_state` values pass through channel reducers (`add_messages` merge in full, append in delta), so replacing reducer values requires `Overwrite` rather than an ordinary update. Full-mode rollback and context compaction replace `messages`; delta resume and delta rollback replace every materialized channel and reset current-head-only channels to their schema default (or `None`). These writes go through `build_state_mutation_graph(as_node, mode, state_schema)`, and `state_schema` MUST be the thread's effective schema (`graph_state_schema(assistant_graph)`), because the base-ThreadState fallback silently discards written channels contributed by custom `AgentMiddleware.state_schema`. Channels absent from a full-mode fork write inherit the parent's channel blobs, so middleware channels survive rollback/compaction (locked by `test_rollback_preserves_middleware_contributed_channels` and `test_compact_thread_context_preserves_middleware_contributed_channels`). The compiled mutation graph has one no-op node (entry = finish) whose checkpoint machinery (channels/versions/metadata) is identical to the agent graph's but schedules no pending tasks, so the restored/compacted head stays idle instead of re-triggering the agent. Never hand-write checkpoints via `checkpointer.aput` for this; raw writers elsewhere must preserve checkpoint parentage — severed ancestry breaks delta replay (see `runtime/runs/worker.py` writer parenting and `checkpoint_patches.py`).
20 
21**Run rollback flow** (`runtime/runs/worker.py`): `_capture_rollback_point` materializes the complete pre-run state via the accessor and captures raw `pending_writes` via `aget_tuple` into an immutable `RollbackPoint` before the run starts — capture failure disables rollback (fail-closed), never restores partial state. In `full` mode, cancel-with-rollback forks from the pre-run checkpoint via the mutation graph and inherits non-message channels from that parent. In `delta` mode, forking is unsafe once the cancelled path has attached sibling writes to the pre-run checkpoint, so rollback replaces every captured channel on the current head, using `Overwrite` for reducers and schema defaults for current-head-only channels. Both modes reattach only the captured pre-run pending writes to the restored checkpoint. Edit replay runs (`metadata.replay_kind="edit"`) also restore the pre-run checkpoint on failed, timed-out, or interrupted completion and publish the restored `values` snapshot to the stream before `end`, so clients do not remain on a transient edited branch when the replay did not produce a successful replacement.
22 
23**Where things live**:
24- `runtime/checkpoint_mode.py` — mode + snapshot-frequency freeze, marker injection, delta detection, compatibility gate, both error types
25- `runtime/checkpoint_state.py` — `CheckpointStateAccessor`, `build_state_mutation_graph`, `RollbackPoint`
26- `checkpoint_patches.py` (package root) — checkpoint-machinery patches: delta-history folding for `InMemorySaver` (delegating to the base walk), stable message IDs across materialization, upstream first-write drop fix, and `BinaryOperatorAggregate` unwrapping an `Overwrite` first write into an empty (MISSING) channel — Union-typed reducer channels (`sandbox`/`goal`/`todos`/`promoted`) have no constructible default, so a replace-style write into a fresh branch thread or a never-written channel stored the wrapper literally and crashed the next consumer (#4380; probe-guarded, stands down if upstream fixes it)
27- `agents/thread_state.py` — `ThreadState`/`DeltaThreadState`, `delta_messages_field` / `DELTA_MESSAGES_FIELD` (`DeltaChannel` at the configured `snapshot_frequency`, default 10), schema adaptation helpers
28- `runtime/context_compaction.py` — compaction via accessor + mutation graph (reference consumer)
29- `runtime/checkpoint_cache/` + `runtime/checkpointer/cached_saver.py` — delta-mode checkpoint history cache; checkpoint state reads MUST go through `CheckpointStateAccessor`, and the checkpointer may be a `CachedHistorySaver` wrapper — never rely on concrete saver types
30- Tests: `tests/test_checkpoint_mode.py` (freeze/detect/gate), `tests/test_checkpoint_state.py` (accessor/mutation graph), `tests/test_delta_channel_checkpointers.py` (saver parity), `tests/test_threads_checkpoint_mode.py`, `tests/test_gateway_checkpoint_mode.py` (dual-mode e2e parity), `tests/test_context_compaction.py` (mutation-graph write, no scheduling), `tests/test_run_worker_rollback.py`, `tests/test_cached_history_saver.py` + `tests/test_cached_history_saver_integration.py` (history cache)
31 
32**Checkpoint channel benchmark**: `scripts/benchmark/checkpoint/bench_channels.py`
33runs paired `full`/`delta` message-only StateGraphs in a fresh child process per
34case, using sync `InMemorySaver` or `SqliteSaver` so reducer, serialization, and
35saver costs stay separate from Gateway/async scheduling. It reports deterministic
36correctness digests, write windows/percentiles, warm and graph-rebuilt cold reads,
37logical checkpoint/write bytes, SQLite DB/WAL/SHM footprint, reducer replay time,
38and peak RSS as versioned JSONL. The controller alternates mode order and rejects
39performance data when paired modes materialize different state. Its default 1 GiB
40estimated cumulative full-payload cap skips both modes of an oversized pair when
41`full` is selected, including every delta cadence in a `--snapshot-frequencies`
42sweep; intentional `--modes delta` diagnostics bypass this full-payload cap, so
43size those runs explicitly. Use `--allow-large-cases` only
44on a provisioned machine. Duplicate CSV matrix values are ignored with a warning;
45use `--repetitions` for repeated samples. Summarize paired successful repetitions
46with `scripts/benchmark/checkpoint/summarize_channels.py` (all ratios are
47`delta/full`). `--profile-dir /tmp/checkpoint-profiles` writes one cProfile
48artifact per case for attribution. Profiled rows carry `profiled: true`, and the
49summarizer automatically excludes them from baseline summaries with a warning.
50Storage-size collection relies on saver-specific diagnostic layouts; if those
51layouts change, the timing/correctness row remains successful while storage
52fields become `null` and `storage_stats_error` records the diagnostic failure.
53Example:
54 
55```bash
56cd backend
57PYTHONPATH=. uv run python scripts/benchmark/checkpoint/bench_channels.py \
58  --backends sqlite --updates 100,500,999,1000,1001 --payload-bytes 128 \
59  --repetitions 7 --output /tmp/checkpoint-bench.jsonl
60PYTHONPATH=. uv run python scripts/benchmark/checkpoint/summarize_channels.py \
61  /tmp/checkpoint-bench.jsonl
62```
63 
64The production-shaped layer lives in
65`scripts/benchmark/checkpoint/bench_production.py`: per-case child processes
66run graph-level `ainvoke` turns through the real lead-agent graph (scripted
67deterministic model, real `AsyncSqliteSaver`), then measure
68`GET /threads/{id}/state` and `POST /threads/{id}/history` through the real
69Gateway route stack in the same event loop (httpx ASGITransport), split into
70cold/warm accessor-graph-cache samples. It sweeps `snapshot_frequency`
71(config: `checkpoint_delta.snapshot_frequency`, process-frozen like the mode),
72pairs every delta frequency against the same full row, and fails both
73rows of a pair when materialized or wire digests diverge. Each case must have
74more than the two discarded warm-up turns, and SQLite DB/WAL/SHM sizes are
75captured while the saver is still open so they represent the online storage
76footprint. Summarize with
77`scripts/benchmark/checkpoint/summarize_production.py` (ratios are
78`delta/full`; it also emits `snapshot_write_spike` and `cache_effect_ms`,
79the decision inputs for the production snapshot-frequency and accessor-cache
80defaults). Harness tests live in `tests/test_bench_checkpoint_production.py`
81and `tests/test_summarize_checkpoint_production.py`; timing thresholds are
82not CI gates. The matrix test pins that every `(repetition, turns)` group
83contains both modes and that their execution order flips between consecutive
84groups, including across repetition boundaries.
85 
86Operational limits learned from the first runs (the default matrix is too
87large to run blindly):
88 
89- The default `--timeout-seconds 900` is insufficient for delta mode at
90  `snapshot_frequency=1000` once turns reach 500 (measured: delta-500 takes
91  ~1100-1200s; delta-2000 takes ~45min). Pass an explicit
92  `--timeout-seconds` for any large matrix, and treat the turns=2000 corner
93  as practical only at small snapshot frequencies.
94- Full-mode 2000-turn runs produce a ~33GB sqlite DB. Point `TMPDIR` at real
95  disk, not tmpfs (the benchmark uses `tempfile.TemporaryDirectory`, which
96  honors `TMPDIR`), or the run dies mid-case.
97- The history route clamps `limit` to 100 (`le=100` on
98  `ThreadHistoryRequest.limit`), so `--history-limits` values above 100 are
99  measured and reported by their effective (clamped) limit.
100 
101Example:
102 
103```bash
104cd backend
105PYTHONPATH=. uv run python scripts/benchmark/checkpoint/bench_production.py \
106  --turns 10,100,500,1000,2000 --payload-bytes 128 \
107  --snapshot-frequencies 10,50,100,500,1000 \
108  --repetitions 7 --output /tmp/production-bench.jsonl
109PYTHONPATH=. uv run python scripts/benchmark/checkpoint/summarize_production.py \
110  /tmp/production-bench.jsonl
111```
112