Continues the zh/en README drift follow-up (#5222 covered the LangGraph
Studio section). The English README documents Request Trace Correlation
as a standalone section between the slash-command table and LangSmith
Tracing; the zh README only mentioned the correlation id in passing
inside its Langfuse section.
Add the translated section in the same position (after the
slash-command table, before LangSmith 链路追踪), covering: X-Trace-Id
inheritance/generation and the response header, propagation to detached
runs / subagents / background memory threads, the logging.enhance
config, the deerflow_trace_id semantics (not a run id, not a provider
trace id, not a lookup key, ignored-and-overwritten when supplied by the
caller), terminal run.delivery receipts with orphan-recovery behavior,
and the loop-detection / deferred-tool-promotion run event records.
The Chinese README had drifted from the English one: the entire
"LangGraph Studio (Optional)" section (including the standalone
langgraph dev workflow, server-owned assistant provenance in that mode,
the persisted-store repair note, the authenticated-identity consumption
for custom agents/skills/uploads/memory, the /mnt/user-data/outputs
native-delivery enforcement, and the dual LangGraph streaming interfaces
for custom events) was never translated and had no zh counterpart.
Add the translated section in the same position it occupies in the
English README (after the local development walkthrough, before
进阶配置), keeping the en section structure aligned. Verified the
section heading now exists on both sides and that no relative links are
affected.
TestBashExecutionHarvest's shell-persistence tests report
shell_persistent=None whenever deerflow.sandbox.overwrite is not already
cached in sys.modules. _setup_executor_classes replaces the
"deerflow.sandbox" parent package with a MagicMock, so a later
`from deerflow.sandbox.overwrite import unwrap_sandbox` inside
_harvest_shell_persistence can no longer locate the submodule through
the mocked parent ("'deerflow.sandbox' is not a package") when the leaf
module is not already in sys.modules. The helper's
`except Exception: return None` silently converts that ImportError into
an UNKNOWN provenance stamp.
Whether the leaf module was cached depended on which tests ran earlier
in the session, making the outcome order-dependent: green in CI by
collection-order luck, red when the module runs alone or first.
Fix it the same way the fixture already handles audit_context and
tool_search: preload the real leaf modules (deerflow.sandbox.sandbox_provider
and deerflow.sandbox.overwrite) before installing the mocked parent
package, pin them in sys.modules for the duration of the test, and
restore the previous state afterwards.
* fix(tests): make three backend test modules runnable on Windows hosts
Follow-up to #5210 (clock-granularity fix) clearing the remaining
deterministic Windows failures in modules that are otherwise
platform-neutral. Five tests fail on Windows for root causes unrelated
to the behavior under test:
- test_pnpm_script.py::test_make_install_dry_run_does_not_invoke_bare_pnpm
shells out to `make`, which Git Bash on Windows does not bundle
(the same gap reported in #5177). Skip when make is unavailable.
- test_mcp_session_pool.py: one test asserts the injected MCP temp dir
has POSIX mode 0o700; Windows has no POSIX mode bits (ntfs reports
0o777), so the mode check now runs only on POSIX. A second test
asserted `TMP.endswith("mcp-internal/tmp")` while Windows tmp paths
use backslashes; normalize the separator before comparing.
- test_skillscan_native.py: two tests build a 3000-operand `1+1+...`
chain to exercise deep-AST resilience. CPython's C recursion limit
for ast construction is platform-dependent (~800 on Windows vs
~8000 elsewhere), so 3000 reliably overflows on Windows and the
scanner records an error instead of findings. 600 chained BinOps
stays deep for the client-analysis walk while fitting the limit on
every supported platform.
No product code is touched; on POSIX the suite behaves exactly as
before.
* fix(tests): make three backend test modules runnable on Windows hosts
Follow-up to #5210 (clock-granularity fix) clearing the remaining
deterministic Windows failures in modules that are otherwise
platform-neutral. Five tests fail on Windows for root causes unrelated
to the behavior under test:
- test_pnpm_script.py::test_make_install_dry_run_does_not_invoke_bare_pnpm
shells out to `make`, which Git Bash on Windows does not bundle
(the same gap reported in #5177). Skip when make is unavailable.
- test_mcp_session_pool.py: one test asserts the injected MCP temp dir
has POSIX mode 0o700; Windows has no POSIX mode bits (ntfs reports
0o777), so the mode check now runs only on POSIX. A second test
asserted `TMP.endswith("mcp-internal/tmp")` while Windows tmp paths
use backslashes; normalize the separator before comparing.
- test_skillscan_native.py: two tests build a 3000-operand `1+1+...`
chain to exercise deep-AST resilience. CPython's C recursion limit
for ast construction is platform-dependent (~800 on Windows vs
~8000 elsewhere), so 3000 reliably overflows on Windows and the
scanner records an error instead of findings. 600 chained BinOps
stays deep for the client-analysis walk while fitting the limit on
every supported platform.
No product code is touched; on POSIX the suite behaves exactly as
before.
Update: address review feedback (P2, recursion-recovery regression)
The 600-operand chain no longer exercises recursion exhaustion on POSIX,
so the recovery handler in _scan_python was unprotected by the renamed
test. Replace the input-based variant with a controlled RecursionError
injected via monkeypatched _find_client_handle_sink (platform-
independent); removing the handler now turns the test red again.
* fix(tests): make three backend test modules runnable on Windows hosts
Follow-up to #5210 (clock-granularity fix) clearing the remaining
deterministic Windows failures in modules that are otherwise
platform-neutral. Five tests fail on Windows for root causes unrelated
to the behavior under test:
- test_pnpm_script.py::test_make_install_dry_run_does_not_invoke_bare_pnpm
shells out to `make`, which Git Bash on Windows does not bundle
(the same gap reported in #5177). Skip when make is unavailable.
- test_mcp_session_pool.py: one test asserts the injected MCP temp dir
has POSIX mode 0o700; Windows has no POSIX mode bits (ntfs reports
0o777), so the mode check now runs only on POSIX. A second test
asserted `TMP.endswith("mcp-internal/tmp")` while Windows tmp paths
use backslashes; normalize the separator before comparing.
- test_skillscan_native.py: two tests build a 3000-operand `1+1+...`
chain to exercise deep-AST resilience. CPython's C recursion limit
for ast construction is platform-dependent (~800 on Windows vs
~8000 elsewhere), so 3000 reliably overflows on Windows and the
scanner records an error instead of findings. 600 chained BinOps
stays deep for the client-analysis walk while fitting the limit on
every supported platform.
No product code is touched; on POSIX the suite behaves exactly as
before.
Update: address review feedback (P2, recursion-recovery regression)
The 600-operand chain no longer exercises recursion exhaustion on POSIX,
so the recovery handler in _scan_python was unprotected by the renamed
test. Replace the input-based variant with a controlled RecursionError
injected via monkeypatched _find_client_handle_sink (platform-
independent); removing the handler now turns the test red again.
Update: address second review feedback (P2, early-stop regression coverage)
The 600-operand tail no longer proves the walk stops after finding a
sink (it completes inside POSIX recursion limits either way). Replace it
with the suggested instrumentation: a sentinel os.system call after the
sink plus an instrumented _walk_client_scope that records any visit to
the sentinel while analysis.found is already set, failing the test if
traversal continues past the sink. Platform-independent; the sentinel's
shell-exec finding comes from the deterministic ast.walk pass and is
irrelevant to the walk guard.
* fix(journal): dedup llm.ai.response persistence on re-fired on_llm_end
LangChain may deliver on_llm_end more than once for the same run_id.
RunJournal already dedups token accounting and the run summary
(_record_message_summary) on that premise via _counted_message_llm_run_ids,
but the durable llm.ai.response self._put() call was left unguarded.
The event store is append-only and count_messages/list_messages read raw
rows without read-time dedup, so a replayed callback persists a second
llm.ai.response row for one logical response while the run's own
message_count counts it once. This inflates count_messages, duplicates a
message in list_messages pagination, and leaves the durable feed
inconsistent with the run summary.
Gate the persistence + summary block by the existing per-run_id guard so a
replayed callback is a no-op, keeping the durable message feed and the run
summary in agreement. Distinct run_ids are unaffected.
Adds regression tests: a re-fired callback for one run_id persists exactly
one row (red on main), and distinct run_ids each still persist a message.
* fix(journal): preserve canonical response on late usage
* fix(journal): preserve late usage while deduplicating responses
* fix(journal): keep first callback response canonical
* fix(journal): snapshot canonical response summaries
---------
Co-authored-by: CorgiBoyG <CorgiBoyG@users.noreply.github.com>
test_list_by_thread creates two runs back-to-back and expects the newer
one to sort first under list_by_thread's newest-first ordering. That
relies on the wall clock advancing between the two create() calls.
On Windows, datetime.now() has a coarse granularity (~15.6 ms), so both
runs can receive an identical created_at. Python's stable sort then
keeps insertion order and the assertion fails; on this host 50
consecutive now_iso() calls return identical strings.
Drive the clock with a controlled 1 ms-per-call fake (same monkeypatch
pattern as test_list_by_thread_is_stable_when_timestamps_tie) so the
strictly-newer assumption no longer depends on the host clock. The tie
case stays covered by the existing dedicated test.
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(lark): enforce private ACLs on Windows credential tree
On Windows, posix chmod(0o700/0o600) does not map to NTFS ACLs, so the
secret-bearing Lark CLI credential tree was not actually owner-restricted
and existing trees were not repaired.
Branch the permission application by platform:
- POSIX: directories 0o700, files 0o600 (behavior unchanged).
- Windows: disable inherited ACLs, grant the Gateway process user Full
Control (resolved via its SID from whoami /user /fo csv /nh so it is
locale-independent), and remove broad non-administrative grants
(Everyone, Authenticated Users, Users). Fail closed on identity or
icacls failures so a tree is never left accessible silently.
Existing-tree handling is covered by asserting every entry in the tree is
repaired, and the Windows command contract is covered by mocked tests run
in CI.
* fix(lark): harden Windows credential tree against TOCTOU and hard-link races
This replaces the path-based Windows hardening (lstat -> SetFileSecurityW(path) -> iterdir) with a handle-relative walker, so validation, the ACL update, and traversal are bound to the opened object rather than a re-resolved pathname.
Every credential object is opened no-follow; children are enumerated with GetFileInformationByHandleEx(FileFullDirectoryInfo) and opened/created relative to an already-open parent handle (NtOpenFile/NtCreateFile with OBJECT_ATTRIBUTES.RootDirectory), so a pathname swap cannot redirect the walk. Credential directories are opened exclusively (share=0): SetSecurityInfo therefore does not propagate the final owner-only OI|CI DACL into as-yet-unvalidated children, and the namespace is locked for the duration of the walk (concurrent child rename/replacement and hard-link insertion fail with sharing violations). Any file with nNumberOfLinks != 1 is rejected before its security descriptor is touched, so an NTFS hard link to an external file cannot change that file owner/DACL. POSIX keeps the lstat-before-descent walk.
Tests: native regressions for exclusive no-propagation, late hard-link insertion being blocked, mid-walk junction swap being blocked, static hard-link rejection, and both real NTFS junction rejections. Mock seams updated for the handle-relative API, and Windows portability fixes make the suite green on Windows except the known #5116 sandbox-runtime executable-bit failures.
* test(lark): keep the credential-tree symlink assertion portable
The credential-tree symlink rejection is a ValueError; POSIX reports a symlink while the Windows handle-relative walker reports a reparse point. Use a platform-dependent regex so the test passes on Linux/macOS and Windows.
* fix(lark): close remaining credential-tree hardening gaps
Review follow-up for the handle-relative credential-tree walker:
- Stage the transaction snapshot under the owner-only root, copying only config/ and data/.
- Serialize ensure() per-user across threads and processes with a dedicated lock.
- Make the walker iterative so deep trees cannot hit the recursion limit.
- Re-reject a symlinked POSIX root before mkdir; drop the over-strict ancestor-chain check.
- Soften the SetSecurityInfo failure claim; add regressions for each and carry os.SEEK_END in the os stub.
* fix(lark): anchor hardening lock under trusted base and keep POSIX untouched
Follow-up refinements to the credential-tree hardening:
- The per-user hardening lock file now lives directly under the trusted base_dir
instead of the unverified per-user chain, so it is never written through an
ancestor that has not yet passed reparse validation.
- ensure() takes the hardening lock only on the Windows branch; POSIX keeps the
original contract, so no new lock-file side effect.
- Strengthen the ancestor-junction regression (lock not written to the external
target) and fix two test docstrings to match the parent-first order and the
no-prior-broadening failure claim.
* fix(lark): anchor credential-operation lock under trusted base on Windows
The per-user credential lock (_lark_credential_lock) created its advisory lock file
under the unverified per-user chain (users/<id>/integrations/.lark-cli.credentials.lock)
before ensure() validated the ancestor chain. On Windows it is now anchored directly
under the trusted paths.base_dir (mirroring the hardening lock), so a junction at
integrations can no longer cause the credential lock to be written into an external
target before reparse validation. POSIX keeps the original location unchanged.
Tests:
- Public-flow regression (start_lark_config -> credential lock -> ensure) uses an empty
sentinel lock file to prove the old credential-lock path is never opened/written.
- CLI-write re-harden tests restore the POSIX outcome assertion (file tightened to 0600).
In tool-enabled requests MindIEChatModel._astream falls back to awaiting the full _agenerate response and re-emitting it as simulated AIMessageChunks. The full response carries usage_metadata, but none of the simulated chunks copied it, so chunk aggregation (add_ai_message_chunks) produced a final message with usage_metadata=None. Token usage therefore vanished from token accounting, run stats, persistence and the UI for every tool-enabled streamed turn.
Mirror OpenAI's terminal-usage-frame convention: attach msg.usage_metadata to exactly the last simulated chunk (the trailing tool-call chunk when present, else the last text chunk / the single tool-only chunk) so the aggregated message carries it exactly once. add_usage() is per-chunk additive, so attaching usage to every chunk would multiply the totals.
Scope: MindIEChatModel only; other providers keep native streaming and ainvoke/non-tool astream were already correct.
Tests: regression guard asserting exactly one carrier chunk equals the last one and that merged usage equals the original across all three simulated-stream branches, plus chain-level tests driving the public astream() wrapper and asserting the persisted model_dump() shape.
Closes#5192
* fix(agents): make the injected current-date timezone configurable
## Why
The date reminder injected into the lead and subagent prompts (DynamicContextMiddleware / SubagentDateContextMiddleware) was formatted with the server's local wall clock. DeerFlow containers default to UTC, so a user in Asia/Shanghai chatting in the 00:00-08:00 window was told that 'today' is the previous day - the model then reasons, plans, and date-stamps against the wrong day.
## What changed
- _format_current_date() now reads the optional DEER_FLOW_DATE_TIMEZONE env var (IANA name, e.g. Asia/Shanghai) and renders the date in that zone.
- Unset = unchanged server-local behavior; invalid names log a warning and fall back to server-local.
- Documented the knob in config.example.yaml, the module docstring, and the DynamicContext entry in agents/middlewares/AGENTS.md.
## Surface area
- [x] Agents / LangGraph - prompt-layer date context only; message shape and midnight-update behavior unchanged
- [ ] Frontend UI / Backend API / Sandbox / Skills / Dependencies
- [x] Default behavior change (opt-in via env var - no behavior change unless set)
## Bug fix verification
- New tests: test_format_current_date_honors_configured_timezone (UTC 20:30 -> 2026-09-03 in Asia/Shanghai), test_format_current_date_defaults_to_server_local_without_env, test_format_current_date_invalid_timezone_falls_back.
- Existing mocked-datetime tests pass unchanged (no env -> datetime.now() path).
## Validation
- cd backend && python -m pytest tests/test_dynamic_context_middleware.py: 31 passed.
- blocking_io/test_dynamic_context_middleware.py: 2 pre-existing abefore_agent failures reproduce identically on clean main (blockbuster os.listdir detection on this host); the other 2 pass.
- ruff format + ruff check clean.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** analysis, implementation, and regression tests produced with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(agents): avoid passing tz to datetime.now when no timezone is configured
CI (backend-unit-tests shard 2) failed in test_tool_error_handling_middleware.py::test_subagent_chain_injects_date_without_memory_and_coalesces_for_strict_provider because its _FrozenDateTime.now() subclass override accepts no arguments, while _format_current_date() called datetime.now(None) even when DEER_FLOW_DATE_TIMEZONE was unset.
- _format_current_date() now calls datetime.now() with no arguments unless a timezone is actually configured, preserving the exact legacy call shape for every datetime-subclass test fake.
- The configured-zone path still calls datetime.now(tz) and converts via astimezone(tz).
- Updated the no-env unit test to assert datetime.now() is called without arguments.
Validation: python -m pytest tests/test_dynamic_context_middleware.py + the previously failing strict-provider test: 32 passed. ruff clean.
* fix(agents): declare the effective current-date timezone in the assembly descriptor
## Why
Maintainer review on the DEER_FLOW_DATE_TIMEZONE change (#5154): the knob is
prompt-affecting, yet both DynamicContextMiddleware and SubagentDateContextMiddleware
were invisible to the agent assembly descriptor - describe_middleware() fell back to
{"probed": true} for unset, UTC, and Asia/Shanghai alike, so deployments that inject
different dates shared one assembly fingerprint and release observers could not
distinguish or audit the behavior change.
## What changed
- Both middlewares now implement release_policy_parameters() -> dict[str, object],
declaring {"current_date_timezone": <name>} as required by the module's middleware
self-description contract.
- The declared value is the normalized effective zone: a configured, valid
DEER_FLOW_DATE_TIMEZONE is reported by its IANA key (ZoneInfo.key); otherwise the
server-local zone is resolved to its IANA key when the platform exposes one and to
its tzname label otherwise (fixed-offset hosts), with "UTC" as the final fallback.
- Added both middlewares to _MIDDLEWARE_DECLARATIONS in
backend/tests/test_middleware_release_policy.py so the existence check and the
construct-and-canonical-hash check cover them.
## Verification
- New tests: test_date_middlewares_declare_configured_timezone (Asia/Shanghai),
test_date_middlewares_declare_utc_timezone, plus resolved-server-local assertions
for the unset and invalid-env paths; both middlewares agree in every case.
- cd backend && python -m pytest tests/test_dynamic_context_middleware.py
tests/test_middleware_release_policy.py: 70 passed.
- Regression spot-check: tests/test_agent_assembly_descriptor.py,
tests/test_tool_error_handling_middleware.py, tests/test_system_message_coalescing_middleware.py:
102 passed.
- ruff check + ruff format clean.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** analysis, implementation, and regression tests produced with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(agents): stabilize the declared date timezone and simplify the formatting path
## Why
Follow-up review on #5154 (willem-bd). The release-policy declaration added in
884cec4b resolved the observability gap but pinned far less identity than its
docstrings claimed, and the formatting path carried a production no-op.
## What changed
- The declared label is now stable and unambiguous: a configured, valid
DEER_FLOW_DATE_TIMEZONE is reported by its IANA key; without one, the
server-local zone is resolved to a real IANA key from the TZ env var or the
/etc/localtime symlink (Linux/macOS); when no key is recoverable (Windows,
stripped containers) the declaration falls back to a stable
`server-local(+-HH:MM)` sentinel carrying the current UTC offset. It never
reports a bare abbreviation - datetime.now().astimezone() yields only a
fixed-offset timezone whose tzname (e.g. CST, EST/EDT, CET/CEST) is
ambiguous or DST-churns, which the assembly descriptor docstring says must
not happen.
- Dropped the redundant astimezone(tz) in _format_current_date():
datetime.now(tz) already returns the instant expressed in tz. The
configured-zone test now fakes datetime.now(tz) semantics (the fixed instant
converted into the requested zone) instead of relying on that conversion.
- Documented why the knob is an env var, not a config-schema field: it is read
at runtime by both date-context middlewares so an operator can point a
container at another zone without mounting a config.yaml (module docstring +
config.example.yaml note).
- AGENTS.md: fixed the glued DynamicContext sentence (missing separator).
- Added tzdata>=2025.1 to the harness runtime dependencies (with uv.lock) so
ZoneInfo works on stripped containers / Windows without an OS zone database.
## Verification
- New tests: test_server_local_timezone_name_reads_tz_env,
test_effective_timezone_sentinel_uses_offset_when_local_zone_is_not_resolvable;
reworked test_format_current_date_honors_configured_timezone to exercise the
real datetime.now(tz) path.
- cd backend && python -m pytest tests/test_dynamic_context_middleware.py
tests/test_middleware_release_policy.py tests/test_agent_assembly_descriptor.py
tests/test_tool_error_handling_middleware.py: 140 passed.
- ruff check + ruff format clean.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** analysis, implementation, and regression tests produced with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(agents): offload subagent date injection off the event loop
## Why
Follow-up review on #5154 (willem-bd, P2): SubagentDateContextMiddleware.abefore_agent()
called _inject() directly, so enabling DEER_FLOW_DATE_TIMEZONE could synchronously
read the OS timezone database (or the tzdata wheel) on a cold cache - filesystem
work on the async subagent execution path whenever no assembly observer resolved the
zone first.
## What changed
- SubagentDateContextMiddleware.abefore_agent() now offloads the injection via
asyncio.to_thread with the same bounded timeout DynamicContextMiddleware uses
(issue #3402); on timeout it logs and skips the date update for that run instead
of blocking the loop.
- Narrowed the exception handling in _date_timezone() and the TZ-env branch of
_server_local_timezone_name() to configuration-shaped failures
(ZoneInfoNotFoundError / ValueError / OSError). Previously a blanket
`except Exception` also swallowed BlockingError raised by the blocking-I/O
regression gate, mislabeling a loop-blocking call as an invalid timezone and
silently degrading to server-local - which made the new regression anchor
useless. Other exceptions now propagate.
## Verification
- New blocking-I/O regression anchor
(backend/tests/blocking_io/test_subagent_date_context_middleware.py): drives a
real create_agent graph under the strict Blockbuster gate with the knob enabled
and asserts the date reminder is injected. Verified it fails (BlockingError) when
the offload is reverted and passes with it in place.
- python -m pytest tests/blocking_io/test_subagent_date_context_middleware.py:
1 passed. The two pre-existing os.listdir failures in
tests/blocking_io/test_dynamic_context_middleware.py reproduce unchanged on this
host (same as clean main).
- python -m pytest tests/test_dynamic_context_middleware.py
tests/test_middleware_release_policy.py tests/test_tool_error_handling_middleware.py
tests/test_agent_assembly_descriptor.py: 139 passed; the single
ToolReceiptMiddleware-ordering failure reproduces with the change stashed
(local extensions registry, unrelated to this PR).
- ruff check + ruff format clean.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** analysis, implementation, and regression tests produced with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(agents): read the direct /etc/localtime symlink target for the zone key
## Why
Follow-up review on #5154 (willem-bd, P2): on macOS, /etc/localtime commonly
points to /var/db/timezone/zoneinfo/<zone>, but Path.resolve() follows that
directory's own symlink and yields a versioned path such as
/private/var/db/timezone/tz/2026c.1.0/zoneinfo/Asia/Shanghai, which matched no
configured prefix. The server-local resolution then returned None and the
assembly descriptor fell back to a server-local(+HH:MM) sentinel even though
the IANA key was available - conflating zones that share an offset and making
DST-based fingerprints unstable.
## What changed
- _server_local_timezone_name() now reads the direct symlink target via
os.readlink("/etc/localtime") instead of Path.resolve(), so macOS' unversioned
zoneinfo path is seen as-is and its IANA key is preserved.
- The zone key is taken from whatever follows the last "/zoneinfo/" segment,
which also handles Apple's canonical versioned path when a direct target
already carries it, and relative targets are normalized against /etc.
- Removed the now-unused Path import and the fixed zoneinfo prefix tuple.
## Verification
- New tests: test_server_local_timezone_name_reads_direct_macos_symlink_target,
test_server_local_timezone_name_reads_apple_versioned_symlink_target, and
test_server_local_timezone_name_normalizes_relative_symlink_target.
- python -m pytest tests/test_dynamic_context_middleware.py
tests/test_middleware_release_policy.py tests/test_agent_assembly_descriptor.py:
105 passed (75 after re-running the first two on the merged main). The blocking
subagent anchor still passes; the two pre-existing os.listdir blocking failures
on this host are unchanged.
- ruff check + ruff format clean.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** analysis, implementation, and regression tests produced with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* bugfix #5179
* test: cover the env-aware _pick_python fallback from #5179
Follow the test_serve_nginx_stop.py extraction pattern: drive the real
_pick_python from serve.sh against a stub-only PATH plus a mocked env.
- python3 succeeds directly but fails through env -> python selected
(red on main, green on this branch)
- env rejects every candidate -> nonzero exit (also red on main)
- healthy PATH with the real env -> python3 preferred, guarding against
over-rejection
MSYS/Git Bash hosts need the stub dir as an MSYS-style (/c/...) PATH
entry, and bash diagnostics may arrive in the console code page, so the
runner decodes output with errors="replace".
* feat(gateway): add /health/ready readiness probe backed by the database
## Why
GET /health only proves the process is up: it returns 200 even when the persistence engine cannot reach the database. Orchestrators already treat it as a readiness gate (docker-compose.yaml marks the gateway service healthy and nginx depends_on service_healthy), so a DB outage or a still-migrating Postgres leaves the stack 'healthy' while every request fails.
## What changed
- New GET /health/ready endpoint: bounded SELECT 1 against the existing persistence engine (deerflow.persistence.engine.get_engine) with a 2s timeout.
- Response is 200 {'status': 'ready', 'database': 'ok'} when reachable, 503 {'status': 'degraded', 'database': 'unreachable'} when the probe fails, and 200 ready with database=not_configured for backend=memory (nothing to probe).
- GET /health is unchanged (pure liveness), and /health/ready is public through the existing /health auth whitelist.
- docker-compose.yaml gateway healthcheck now polls /health/ready so service_healthy reflects database reachability.
- Documented both endpoints in backend/app/gateway/AGENTS.md.
## Surface area
- [x] Backend API - new GET /health/ready endpoint under backend/app/gateway
- [x] Sandbox / Docker - gateway healthcheck in docker/docker-compose.yaml now gates on readiness
- [ ] Frontend UI / Agents / Skills / Dependencies
- [x] Default behavior change - existing /health unchanged; the prod compose healthcheck is stricter (503 while the database is unreachable)
## Validation
- New unit tests in backend/tests/test_gateway_health.py cover ok / unreachable / not_configured probe results and the 200/503 payload mapping (6 passed).
- app.gateway.app imports cleanly and registers both /health and /health/ready.
- ruff check + ruff format clean.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** design, implementation, and unit tests produced with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(helm): point the gateway readiness probe at /health/ready
## Why
Review on #5166 (willem-bd, P1): the chart still probed /health for readiness,
so Kubernetes marked the pod ready and routed traffic while the database was
unreachable - exactly the failure mode /health/ready was added to catch.
## What changed
- deploy/helm/deer-flow/templates/gateway-deployment.yaml: readinessProbe
httpGet.path now hits /health/ready (DB-backed, 503 while the database is
unreachable). The liveness probe stays on /health.
## Verification
- One-line path change inside the existing readinessProbe block; git diff
confirms only the readiness path changed (liveness untouched).
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** implemented the reviewer-requested probe path change; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(gateway): readiness probe also checks the effective checkpointer/Store backend
## Why
Follow-up review on #5166 (willem-bd, P1): get_engine() only represents the
ORM backend selected by `database:`. The legacy `checkpointer:` section takes
precedence for the LangGraph checkpointer and Store, so a split configuration
(a local SQLite/memory `database:` with `checkpointer.type: postgres`) could
report 200 while the PostgreSQL backend agent runs depend on was down.
## What changed
- GET /health/ready now probes both persistence halves: the ORM engine behind
`database:` (unchanged) and the effective LangGraph checkpointer/Store
backend resolved with the runtime's own rule (legacy `checkpointer:` config,
otherwise derived from `database:`), for memory/sqlite/postgres backends.
- The payload gains a `checkpointer` field with the same
ok / not_configured / unreachable vocabulary as `database`; 503 degraded is
returned when either probe is unreachable.
- Probes are bounded by the existing 2s timeout: sqlite via aiosqlite SELECT 1
on the resolved path, postgres via a bounded psycopg AsyncConnection SELECT 1
on the DSN with the configured search_path. A missing driver for a configured
backend degrades readiness (the runtime could not run either).
- Documented the two-probe semantics in the endpoint docstring and
backend/app/gateway/AGENTS.md.
## Verification
- New tests: healthy ORM engine + unreachable legacy checkpointer backend ->
503 degraded with database: ok / checkpointer: unreachable; checkpointer
probe mapping for memory/sqlite(postgres missing-driver) backends; existing
payload tests now pin the checkpointer field.
- cd backend && python -m pytest tests/test_gateway_health.py: 11 passed.
- app.gateway.app imports cleanly; ruff check + ruff format clean.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** design, implementation, and unit tests produced with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(gateway): bound /health/ready to one deadline and probe the startup checkpointer snapshot
## Why
Second round of review on #5166 (zhfeng P1/P2, willem-bd P1/P1/P2). Three
correctness issues remained in the readiness endpoint:
- The database and checkpointer probes ran sequentially, each allowed 2s, so a
healthy response could take almost 4s - past Kubernetes' 1s default
readinessProbe timeout and inside Docker's 3s client timeout. A slow but
healthy backend could make every replica unready.
- The checkpointer probe re-resolved process-wide, hot-reloaded configuration
per request, while app.state.checkpointer/store are built once from the
startup_config snapshot in langgraph_runtime(). After a live config edit the
endpoint could probe a backend the running gateway does not use, and a
resolution failure was swallowed into None -> not_configured -> 200.
- The SQLite probe opened the path with aiosqlite.connect(), which creates the
file when missing: a deleted checkpoint database was silently resurrected as
an empty file and reported ok instead of surfacing the outage.
## What changed
- backend/app/gateway/health.py: the two probes now run concurrently beneath a
single endpoint-wide deadline (_READINESS_DEADLINE_SECONDS=3.0) so a healthy
response completes within one probe window (~2s), never the sum of both.
A probe that overruns the deadline degrades the endpoint instead of hanging.
- langgraph_runtime() now records the checkpointer/Store config resolved from
the same startup_config snapshot its checkpointer/store singletons are built
from (app.state.checkpointer_config); /health/ready probes that snapshot and
never re-resolves hot-reloaded config. resolve_checkpointer_config() returns
None on resolution failure and the endpoint fails closed (503, checkpointer:
unreachable) instead of reporting not_configured.
- The SQLite probe opens disk-backed paths with the non-creating mode=rw URI
flag, so a missing database file stays missing and yields unreachable;
in-memory forms (:memory:, file:...mode=memory) have nothing external to
probe and report not_configured like the memory backend.
- Orchestrator timeouts now sit above the endpoint bound: Helm readinessProbe
gains timeoutSeconds: 5 (Kubernetes default is 1s) and the docker-compose
gateway healthcheck client timeout moves from 3s to 5s.
## Verification
- New regression tests: concurrent probes keep total elapsed time within one
probe window; a probe ignoring its budget trips the endpoint deadline to 503;
missing SQLite file stays absent and yields unreachable; in-memory SQLite
forms map to not_configured; missing startup snapshot / config resolution
failure fail closed to 503; resolve_checkpointer_config() raising is covered.
- cd backend && python -m pytest tests/test_gateway_health.py: 21 passed;
tests/test_gateway_docs_toggle.py and lifespan/shutdown gateway suites pass.
- ruff check + ruff format clean on all changed files.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** implemented the reviewer-requested concurrency/deadline, startup-snapshot probing, fail-closed resolution, and non-creating SQLite probe; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(gateway): serialize connection-opening readiness probes behind a strict gate
## Why
Review on #5166 (willem-bd, P1): every request to /health/ready opened a new
PostgreSQL connection in _probe_postgres_backend, outside both the ORM pool and
the runtime checkpointer pool. The route is public through the /health auth
prefix and nginx proxies /health/*, so concurrent unauthenticated requests
could create an unbounded number of connections (each held for up to two
seconds), exhaust PostgreSQL max_connections, and take down both normal
traffic and the readiness probe itself.
## What changed
- backend/app/gateway/health.py: connection-opening checkpointer probes
(sqlite connect, postgres AsyncConnection.connect) now run inside a strict
per-process gate - an asyncio.Lock cached per running event loop - so at
most one probe connection can be in flight per worker process. Requests
that queue behind the gate are still shed by the existing endpoint-wide
deadline, so a flood cannot pile up new connections or open files.
- Memory and unknown-backend decisions stay outside the gate; payload and
probe semantics are unchanged. The serialization is documented in the
module docstring and backend/app/gateway/AGENTS.md.
## Verification
- New regression test: 8 concurrent readiness_payload() requests against an
instrumented sqlite probe assert the maximum number of in-flight probe
connections is 1 while every request still returns 200.
- cd backend && python -m pytest tests/test_gateway_health.py: 22 passed;
tests/test_gateway_docs_toggle.py and tests/test_gateway_lifespan_shutdown.py
also pass on the merged main head.
- ruff check + ruff format clean on all changed files.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** implemented the reviewer-requested strict concurrency bound for the public readiness probe; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(runtime): reclaim idle keyed locks safely
Replace the per-loop thread lock registries with a waiter-aware keyed lock table. Count holders and queued waiters before acquisition so idle entries can be reclaimed without allowing a late caller to bypass an existing waiter.
Add regression coverage for runtime call-site reclamation, goal/checkpoint domain independence, queued-waiter ordering, cancellation cleanup, high-cardinality key reclamation, and cross-event-loop isolation.
Fixes#5171
* style(runtime): format keyed lock helper
* fix(mcp): keep ToolRuntime injection for sync-wrapped MCP tools
make_sync_tool_wrapper attached an annotation-less wrapper to tool.func,
which made LangGraph's ToolNode stop detecting the coroutine's
"runtime" parameter (_get_all_injected_args falls back to func first
and its type hints are empty). Every MCP tool in a sync agent caller
then ran with runtime=None: resolve_runtime_user_id fell through to the
default user, and the background-submit wrapper lost run_id/tool_call_id
on the TaskSubmitRequest, so completion notifications launched under the
default lead agent instead of the thread's agent.
Wrap the generator and both sync_wrapper variants with functools.wraps
so get_type_hints still sees the original annotations.
Adds a regression test that drives a func-patched pooled MCP tool
through a real ToolNode and asserts the ToolRuntime is injected with
the thread's user context. It fails on main (runtime=None) and passes
with the fix.
* docs(mcp): record sync-wrapper annotation contract; extend regression coverage
Address review feedback on #5164:
- Expand the Notes block in make_sync_tool_wrapper to state the functools.wraps
contract (copies __name__/__qualname__/__doc__/__annotations__/__dict__ and
sets __wrapped__) and why that is what keeps get_type_hints resolving string
annotations from callers like
mcp/tools.py and skill_manage_tool.py. Drop the no-op wraps on the inner
run_coroutine so the wrap surface stays minimal.
- Rename the regression test to test_func_patched_mcp_tool_keeps_toolnode_runtime_injection.
- Add test_sync_wrapped_builtin_tools_still_resolve_runtime to pin that the
built-in tools (which carry runtime as a pydantic schema field) keep resolving
runtime after their func is wrapped by make_sync_tool_wrapper, so a future
wrapper refactor cannot silently regress per-user resolution for them.
* Fixes#5060: P1 snapshot config before loading, P2 RLock for sync path
P1: Config changes during initialization can permanently cache stale tools.
- Snapshot _config_path and _config_signature BEFORE await get_mcp_tools()
- Compare AFTER get_mcp_tools() completes using _current_config_state()
- If config changed during loading, discard stale result and retry
- Prevents publishing old tools with new signature, which would make
_is_cache_stale() permanently return False
P2: Module-level asyncio.Lock still fails across event loops after real contention.
- _init_lock = threading.RLock() for sync path (reentrant, prevents races)
- _async_init_lock = asyncio.Lock() for async init serialization
- reset_mcp_tools_cache() now acquires _init_lock for serialization
Also fixes test P3: removed duplicated test bodies that leaked state between tests.
* fix(mcp): make cache initialization cross-loop safe
- Replace the module-level asyncio.Lock with thread-safe generation claiming
- Snapshot config state before/after MCP loading and discard stale results
- Keep reset state changes short and non-blocking for async endpoints
- Add regression coverage for contended cross-loop init, config rewrites during load, and reset while init is in flight
* fix: release MCP init claim on cancellation
Release the in-flight generation claim from a cancellation-safe finally block so cancelling the task that owns initialization does not strand future callers. Add regression coverage for cancelling the owner and then reinitializing successfully.
* fix(mcp): retire session pool before cache reset release
Prevent a concurrent MCP cache initializer from publishing tool wrappers
bound to the session-pool singleton that reset_mcp_tools_cache() is
already retiring. Add regression coverage for that interleaving.
* fix(mcp): retire session pool on stale cache invalidation
* fix(mcp): retire pool on init discard
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
## Why
The composer in input-box.tsx carried a commented-out legacy <PromptInputActionMenu> attachments block (TODO: Add more connectors here) left over from before the AddAttachmentsButton component replaced it. Dead commented UI misleads maintainers into thinking the old path is live or half-migrated, and it has no runtime effect.
## What changed
- Deleted the 9-line commented-out JSX block between <PromptInputTools> and <AddAttachmentsButton>.
- No component, import, or i18n key changed: PromptInputActionMenu* are still used by the live menus below, and AddAttachmentsButton already provides the attachments entry point.
## Surface area
- [x] Frontend UI - composer tool row, comment-only change
- [ ] Backend API / Agents / Sandbox / Skills / Dependencies / Default behavior change
## Validation
- Comment-only deletion: no behavior change; the surrounding JSX is byte-identical outside the removed lines.
- Full pnpm check requires node_modules install on this host; diff is limited to dead comments so lint/typecheck risk is nil.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** located and removed the dead block with AI assistance; reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* feat(e2b-sandbox): make mount upload deadline configurable
Replace the hardcoded 120-second mount upload deadline with a
configurable `mount_upload_deadline_seconds` key read from
SandboxConfig (extra=allow). The value is validated: zero and
negative inputs are clamped to 1 second. Omitting the key
preserves the existing 120-second default.
This addresses the follow-up from PR #4842 review: operators
with large mounts or slow networks can now size the deadline to
their deployment without changing code.
* fix(e2b-sandbox): address review feedback on configurable deadline
- Remove import-time default capture from _mount_deadline_reason()
and _MountUploadBudget.deadline_seconds to prevent silent drift.
- Add warning log when mount_upload_deadline_seconds is clamped to 1
(was silent before).
- Update AGENTS.md E2B Mount Uploads section: deadline is now
configurable, not fixed 120.
- Add mount_upload_deadline_seconds to YAML examples in provider
docstring and __init__.py.
- Add config-path test that exercises SandboxConfig -> _load_config ->
_apply_mounts end-to-end.
* feat(e2b-sandbox): surface structured mount upload result on sandbox
Introduce MountUploadResult dataclass and attach it to
E2BSandbox.mount_upload_result after creation. This makes mount
truncation observable in code without re-parsing Gateway logs.
_apply_mounts() now returns MountUploadResult with truncated, reason,
and upload totals. _create_sandbox() captures the result, stores it on
the sandbox instance, and records it in a provider-level map so the
result survives warm-pool reclaim and reconnect.
MountUploadResult.truncated is True only when the upload pass was
stopped early by a resource limit (deadline, file count cap, or byte
budget). Individual mount failures (missing host path, SDK errors) are
logged but do NOT set truncated.
Tests cover: success totals, deadline truncation, file-count truncation,
byte-budget truncation, non-limit failure not reported as truncation,
missing host path not reported as truncation, create→sandbox wiring,
and create→release→warm-pool→acquire result preservation.
* fix(e2b-sandbox-provider): fix _mount_results lifecycle leak and review findings
- Add _forget_mount_result() helper and call it at all terminal sandbox
paths: _reuse_in_process_sandbox dead-evict, _reclaim_warm_pool_sandbox
reconnect/dead/bootstrap/ownership/shutdown failure branches,
_forget_local_sandbox, _kill_and_close. Prevents unbounded dict growth
over a long-running Gateway process.
- Make MountUploadResult @dataclass(frozen=True) to prevent silent mutation
of the shared reference between provider map and sandbox attribute.
- Move _mount_results insert under self._lock in _create_sandbox to match
the read discipline in _register_connected_sandbox.
- Guard _resolve_mount_upload_deadline against None (YAML explicit null)
to avoid int(None) TypeError.
- Add 5 regression tests covering each bypass path and the frozen invariant.
* fix(e2b-sandbox-provider): add _forget_mount_result to _evict_oldest_warm branches
Add _forget_mount_result() calls to all four terminal exit paths in the
E2B _evict_oldest_warm override (reconnect failure, already-gone, kill
failure, kill success). The peer-owned path already cleans up via
_forget_local_sandbox. Add test_evict_oldest_warm_cleans_mount_result to
pin the kill-success branch.
* docs: reduce agent guidance size
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
## Why
DeerFlow declares one timestamp convention in deerflow/utils/time.py: every lifecycle timestamp is UTC (now_iso / datetime.now(UTC)). SubagentResult writers in subagents/executor.py still used naive datetime.now(), so on any non-UTC host the in-memory lifecycle metadata (started_at / completed_at) was local wall-clock time. The sibling durable-batch path (subagents/batch_service.py) already stamps datetime.now(UTC), so the same run model carried two different conventions depending on which path wrote it.
## What changed
- Added an executor-local _utcnow() helper that stamps datetime.now(UTC).
- SubagentResult.completed_at default in try_set_terminal(), result.started_at in _aexecute(), and the started_at default in _aexecute_admitted() now route through _utcnow().
- Explicit caller-supplied timestamps (completed_at=...) still pass through unchanged.
- Added regression tests asserting the default writers produce UTC-aware datetimes.
## Surface area
- [x] Backend runtime (deerflow.subagents.executor) - internal dataclass lifecycle metadata, no wire format change
- [ ] Frontend UI / Backend API / Sandbox / Skills / Dependencies / Default behavior change
## Bug fix verification
- New tests: tests/test_subagent_executor.py::test_timestamp_writers_stamp_utc_aware_datetimes and test_utcnow_helper_returns_utc_aware_datetime encode the convention.
- Updated BlockingDateTime.now() in the terminal-publication-order test to mirror datetime.now's optional tz argument.
## Validation
- cd backend && python -m pytest tests/test_subagent_executor.py: 136 passed; 2 pre-existing TestBashExecutionHarvest failures reproduce identically on clean main (Windows sandbox env), unrelated to this change.
- ruff format + ruff check clean on both changed files.
## AI assistance
**Tool(s) used:** Codex (coding agent)
**How you used it:** analysis of the timestamp conventions, implementation, and regression tests authored with AI assistance; change reviewed before commit.
- [ ] I've read and understand every line of this change and take responsibility for it — it's not unreviewed AI output.
* fix(summarization): resolve fraction triggers from declared context_window, degrade instead of crashing the agent build
A fraction trigger/keep clause requires profile["max_input_tokens"], which any
third-party OpenAI-compatible model lacks, so SummarizationMiddleware
construction raised ValueError out of create_summarization_middleware and failed
the whole agent build (#3103).
- factory: translate a declared model context_window into the langchain
profile (metadata-only, never reaches the provider payload); explicit
caller/override profiles win
- summarization factory: drop unusable fraction trigger clauses (absolute
clauses survive), fall a fraction keep back to the messages default, and
disable compaction with an actionable warning only when no usable trigger
clause remains — the agent build never dies from summarization config
- docs: config.example.yaml, ModelConfig.context_window, summarization.md
* refactor(summarization): share the default keep constant with the fraction fallback
The fraction-keep degradation fallback hardcoded ("messages", 20),
duplicating SummarizationConfig.keep's default_factory literal. Move the
value to a shared DEFAULT_KEEP constant so the two cannot drift apart.
* fix(summarization): keep trigger-null + fraction-keep constructing after degradation
A trigger of None with a fraction keep hit the all-clauses-dropped branch
(has_usable_trigger=False) and disabled compaction, and the accompanying
warning claimed configured triggers were all fraction-based when none were
configured. Only report nothing-usable when trigger clauses actually
existed; trigger:null keeps constructing the never-firing middleware with
the degraded keep, matching its behavior outside the degradation path.
* fix(summarization): address review — keep manual compaction, validate ContextSize, pin wiring
Review follow-ups on #4901:
- When every configured trigger is a dropped fraction clause, keep
constructing the never-firing middleware (trigger=None) instead of
returning None: manual /compact runs with force=True and never consults
trigger clauses, so it must keep working for a profile-less model
rather than reporting 'compaction is disabled'. The warning now says
auto-compaction will not fire while manual compaction remains.
- ContextSize gains a config-load validator: fraction values must be in
(0,1] (a percent-style 80 instead of 0.8 previously produced a threshold
the context could never reach — a silently inert trigger), absolute
values must be positive.
- New un-monkeypatched integration test pins the shipped wiring
(context_window declared -> real factory attaches profile -> fraction
clause survives -> middleware constructs), which the stubbed
middleware-side tests and kwarg-capturing factory-side tests each
stopped short of.
- Docs (summarization.md + config.example.yaml) clarify that the fraction
resolves against the summary/anchor model's context_window
(summarization.model_name when set, else the run model), including the
mismatch caveat for a larger-window summary model.
* fix(summarization): reject non-finite ContextSize values at config load
YAML .nan / .inf pass pydantic's float parsing, and nan <= 0 is False,
so the positivity check alone let them through as dead thresholds
(count >= nan is always False) — the same silent-inert-trigger class the
range validator was added to close. Guard with math.isfinite first,
consistent with the existing non-finite guards on mem0 timeout_seconds
and poll_after_seconds.
* fix(summarization): merge context_window into inferred profile, require whole message counts
- construct the model first, then merge max_input_tokens into the
provider-inferred langchain profile: passing profile= to the
constructor replaced the whole inferred metadata (tool_calling,
structured_output, io capabilities, output limits) with the single
key. An explicitly configured profile is still never clobbered.
- reject non-integral ContextSize values for type=messages at config
load: langchain slices the message list with them, so a float index
raised TypeError mid-compaction.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(auth): correct OAuth uniqueness error and index parity on Postgres
create_user() caught any IntegrityError on commit and always reported it
as a duplicate email. The email pre-check already rules out a real email
collision in the common case, so any IntegrityError reaching that handler
is actually idx_users_oauth_identity firing instead -- confirmed against
both backends: SQLite reports "UNIQUE constraint failed:
users.oauth_provider, users.oauth_id", Postgres reports a
UniqueViolationError naming the same index. The caller saw "Email already
registered" for an OAuth account conflict, which is wrong and would send
API consumers debugging the wrong field.
Distinguish the two cases via a substring check on the driver error text
(both backends name the oauth columns) and raise an accurate message for
each.
Also add postgresql_where to the same index, alongside the existing
sqlite_where. This is not a correctness fix -- verified empirically that
Postgres already enforces the same practical uniqueness without it
(NULL is never equal to NULL in either backends unique index, so real
duplicate (provider, id) pairs are already rejected and NULL/NULL rows
are already unconstrained). postgresql_where makes the index genuinely
partial on Postgres too, matching the stated intent in the surrounding
comment and keeping the index smaller as the common case (plain-password
accounts, both columns NULL) accumulates.
* test(bench): add multi-process SQLite vs Postgres concurrency benchmark
CONFIGURATION.md documents that multi-worker deployments must use Postgres
because "SQLite silently ignores row-level locks", but nothing in the repo
exercised that claim against real separate worker processes -- the existing
checkpoint benchmarks (scripts/benchmark/checkpoint/) measure single-process
read/write latency, and the existing Postgres tests
(test_pg_schema_integration.py, test_multi_worker_postgres_gate.py) cover
schema placement and config validation, not throughput or lock behavior
under concurrent load.
run_concurrency_bench.py spawns N real OS processes (subprocess.Popen, not
asyncio tasks or threads within one process) against the shared users
table, mixing reads (get_user_by_email) and writes (create_user) at a
configurable ratio, and reports throughput, error counts by exception
type, and p50/p95/p99/max latency per run.
Measured locally (2/4/8/16 workers, 100 ops/worker, 70/30 read/write):
SQLite completed all operations with zero errors at every worker count
(busy_timeout absorbs contention rather than raising), but total
throughput stayed flat around 28-34 ops/s regardless of worker count, and
p99 latency grew from ~400ms at 2 workers to ~5.9s at 16, with a 22s max.
Postgres throughput scaled with worker count (41 to 66 ops/s) and p99
stayed under 500ms at every worker count tested. Raw JSON output from
both runs is available on request; exact numbers will vary by machine and
are not asserted in the test suite.
test_bench_concurrency.py unit-tests the pure aggregation logic
(percentile math, error grouping, crashed-worker handling) the same way
test_bench_checkpoint_channels.py does for the existing benchmarks --
fast, no DB required, not the full multi-process sweep in CI.
* fix(auth): inspect the driver exception for OAuth conflict detection
str(exc) embeds the full failed INSERT statement, whose column list
always names oauth_provider/oauth_id, so a substring check on it
misclassified every commit-time IntegrityError on the users table as
an OAuth conflict (reproduced on SQLite: a duplicate primary key with
a different email raised "OAuth account already linked: None/None").
_is_oauth_identity_violation now inspects exc.orig instead: constraint_name
on Postgres, both violated column names present (not a bare "oauth"
substring) on SQLite.
Also ships the alembic revision idx_users_oauth_identity's postgresql_where
predicate needed: 0001_baseline created it as a full index on Postgres,
and ORM metadata changes only affect fresh create_all databases, never an
already-versioned deployment.
Addresses review feedback from willem-bd.
* fix(bench): run the concurrency benchmark in an isolated schema and derive paths from the checkout
--pg-url accepted an arbitrary database URL while the code pinned
postgres_schema="public" and unconditionally ran DELETE FROM users --
against any non-disposable database that permanently destroyed every
auth account. Each run now generates a unique throwaway schema
(bench_<uuid>), points both the seeder and every worker subprocess at
it via postgres_schema, and drops only that schema (DROP SCHEMA ...
CASCADE) once the full worker-count sweep finishes.
Also stopped hard-coding /opt/deer-flow/backend as the checkout path
and .venv/bin/python3 as the interpreter: BACKEND_DIR is now derived
from Path(__file__), and workers are spawned with sys.executable (the
orchestrator's own interpreter) instead, so the documented
uv run python scripts/benchmark/concurrency/run_concurrency_bench.py
command works from any checkout.
Addresses review feedback from willem-bd.
* fix: shorten oauth-index revision id, repin migration-head assertions, fix bench read/write mix
- 0017_users_oauth_identity_partial_pg (36 chars) exceeded
alembic_version.version_num's VARCHAR(32) limit, which would fail
stamping/upgrading on both fresh and existing Postgres deployments.
Renamed to 0017_oauth_identity_pg_partial (30 chars).
- Repinned every test asserting 0016_subagent_batches as the migration
head (test_persistence_bootstrap[.py|_concurrency.py|_regression.py],
test_migration_0004/0007/0015) to the new 0017 revision id.
- worker.py's `(i % 100) < int(read_ratio * 100)` assumed n_ops >= 100;
at the documented default (50 ops/worker, 0.7 read ratio) it produced
either all-reads or all-writes, never the claimed mixed workload.
Replaced with read_count()/is_read_op(), which distribute an exact
round(n_ops * read_ratio) reads evenly across the sequence via modular
spacing, and added test_bench_worker.py covering the default values
plus small op counts.
* fix(bench): establish a real physical connection before timing ops
async with sf(): pass entered an empty AsyncSession without checking out
a physical connection -- SQLAlchemy stays lazy until the first statement
executes. That pushed connection-establishment cost onto each worker's
first timed operation instead of conn_time_s, and at 16 workers those 16
cold first-ops (1% of a 1600-op sample) could skew the reported p99.
Execute a real `SELECT 1` before starting the timer instead.
Verified with a real end-to-end run (uv sync + sqlite backend, 2
workers/10 ops, 0 errors) plus the full auth/bench/migration-bootstrap
suites (135 tests) and ruff check/format, all clean.
* fix(bench): synchronize workers before timing, fix percentile off-by-one
Two remaining measurement issues from review:
- run_workers() started the wall clock before spawning any worker, so
throughput/wall_time absorbed N processes' staggered Python-startup and
connection-establishment cost, and early workers could run ahead of ones
still starting. Workers now print READY right before their timed loop
and block on stdin for a GO signal; the orchestrator waits for every
READY, then starts the timer and releases all workers together.
- summarize()'s pct() used int(len(latencies) * p) directly as a
zero-based index -- a one-based-rank-as-index bug that put p95 and p99
at the same slot (the max) for any 20-or-fewer-sample run, and for the
documented 100-sample default. Now delegates to
checkpoint_bench_common.percentile(), the already-correct nearest-rank
implementation used elsewhere in the same benchmark family, instead of
a second, broken one.
Verified: 14/14 unit tests pass (2 new pinned-value regression tests for
the percentile bug, using the reviewer's own 20-sample repro), ruff
clean, and a real 2/4-worker SQLite multi-process smoke run completes
with distinct p95/p99/max latencies and no hang.
* fix(bench): absolute SQLite bench path, surface crash diagnostics, exit nonzero on failure; share OAuth index constant + cover Postgres branch
Three more findings from review at 5fd25a7:
- seed_baseline() cleaned an absolute .deer-flow/bench_data path but
handed DatabaseConfig a relative one, which resolves against the
CALLER's CWD -- not BACKEND_DIR. Invoking the documented command from
anywhere other than backend/ silently pointed the seeder and the
(cwd=BACKEND_DIR) workers at two different directories: workers crashed
with 'unable to open database file' while the run still printed a
well-formed summary and exited 0. Both seed_baseline() and worker.py's
make_session_factory() now use the same absolute path.
- Crashed workers' stderr was captured then discarded, and main() always
exited 0 -- an all-crashed sweep was indistinguishable from a real
(uneventful) measurement to anything checking the exit code or
--out. run_workers() now prints each crash immediately and tags it with
the real worker_id (previously always None); summarize() exposes
crashed_worker_errors alongside the existing crashed_workers count;
main() exits 1 via the new summary_indicates_failure() whenever any
sweep crashed or fell short of expected_total_ops.
- idx_users_oauth_identity was hardcoded separately in the ORM Index and
in _is_oauth_identity_violation's Postgres branch, with no test to
catch drift, and that branch had zero non-skipped coverage (its only
guard needs a live Postgres CI never configures). Exported
OAUTH_IDENTITY_INDEX_NAME from user/model.py as the shared source of
truth (migrations intentionally keep their own frozen literal, matching
every other revision in that package) and added stub-exception unit
tests pinning both the asyncpg constraint_name path and the sqlite
message-substring path, positive and negative.
Verified: 107 passed locally (auth + bench-unit suites), ruff clean, and
two real reproductions -- invoking run_concurrency_bench.py from a
scratch directory outside backend/ (the reviewer's exact repro) now
completes 8/8 ops with crashed_workers: 0 instead of crashing, and the
new crashed_worker_errors/exit-code logic is exercised directly by the
new unit tests against the real summarize()/summary_indicates_failure().
* fix(auth): attribute create_user IntegrityErrors to the right constraint
Two coupled review findings on the classification helpers:
P3 (fall-through) -- after ruling out the OAuth-identity index, create_user
raised "Email already registered: {email}" for every remaining
IntegrityError, including the duplicate-primary-key case the new
regression test exercises, whose address is not registered. Added
_is_email_violation() so the email message is used only for an actual
users.email collision that raced past the pre-check; anything else (in
practice a duplicate id) now raises a neutral
"User already exists (constraint: <name>)".
P2 (unreachable asyncpg branch) -- exc.orig is not the asyncpg error.
SQLAlchemy's asyncpg dialect re-raises its own DBAPI IntegrityError
(pgcode/sqlstate only) 'from' the real asyncpg error, so constraint_name
lives on exc.orig.__cause__. getattr(exc.orig, "constraint_name", None)
was always None on Postgres; the helpers only worked there by accident,
matching asyncpg's DETAIL line in the message fallback. Added
_driver_constraint_name() which walks orig then orig.__cause__, and the
stub tests now model that real shape (orig wrapper + __cause__) instead of
a constraint_name that no driver puts on orig directly.
Tests: 76 passed. New coverage for the email-race path, both new helpers
on each backend, the neutral fallback message, and the cause-chain walk.
* fix(bench): match app SQLite PRAGMAs in workers; fail a sweep on any op error
Two review follow-ups:
- worker.py opened its SQLite engine with only connect_args timeout=30.
synchronous and foreign_keys are per-connection PRAGMAs, so workers ran
at SQLite's synchronous=FULL / foreign_keys=OFF while a real Gateway
worker runs synchronous=NORMAL (persistence/engine.py::_enable_sqlite_wal)
-- an extra fsync per commit on the measured 30%-write path, overstating
SQLite's cost in the direction that flatters the "use Postgres"
conclusion. Added a connect listener applying the same four PRAGMAs, with
a test asserting synchronous/foreign_keys/journal_mode on a real worker
connection.
- summary_indicates_failure() only looked at crashes and the completed vs
expected op counts, so a sweep where every op completed but raised
(e.g. writes hitting OperationalError) passed as a clean measurement:
completed_ops == expected, 0 crashes. Added an "errors > 0" clause; the
error breakdown stays in the JSON, only the exit code changes. Test added.
test_bench_concurrency.py + test_bench_worker.py green (20), plus a real
2-worker sqlite smoke run (6/6 ops, 0 errors, exit 0).
* fix(auth): match the real email index name; only claim "exists" for uniqueness
Review follow-ups on the classification helpers:
- email is mapped_column(unique=True, index=True), which SQLAlchemy and
0001_baseline realise as a single UNIQUE INDEX (ix_users_email), not a
named UNIQUE constraint. _is_email_violation compared the driver
constraint name against "users_email_key", which Postgres never emits,
so that arm was dead on Postgres (SQLite matched via the message). Fixed
to ix_users_email.
- the residual IntegrityError fallback raised "User already exists" for
every remaining IntegrityError -- a NOT NULL / CHECK / foreign-key
violation is not a "user already exists" condition and is not part of
create_user's ValueError contract. Added _is_uniqueness_violation
(sqlstate 23505, or the SQLite "UNIQUE/PRIMARY KEY constraint failed"
message); only that raises the "already exists" ValueError, everything
else propagates unchanged.
- documented scripts/benchmark/concurrency/ in backend/AGENTS.md alongside
the other benchmark family.
Tests: 78 auth + 20 bench-unit pass, ruff clean. New coverage for
_is_uniqueness_violation on both backends and for a non-uniqueness
IntegrityError propagating out of create_user.
* fix(bench): don't pre-close worker stdin (breaks communicate); require --pg-url for postgres
* fix(bench): ruff format; time throughput on the op phase, not teardown
- lint-backend: ruff format the files touched in this PR.
- Throughput window (P2): the orchestrator sampled its wall clock after
every worker's communicate() returned, so it also covered each worker's
engine.dispose(), result serialization and stdout transfer. Each worker
now times just its operation phase (GO -> last op) and reports
ops_elapsed_s; summarize() uses max(ops_elapsed_s) over the workers -- all
released by the same GO -- as the throughput window (ops_window_s).
- Exercise migration 0018 (P2): test_user_oauth_partial_index.py goes
through bootstrap create_all(), which builds the partial index from ORM
metadata and never runs 0018.upgrade(). New Postgres-gated
test_migration_0018_oauth_identity_pg_partial.py alembic-upgrades to 0017
(full index), then 0018 (asserts the predicate appears), then downgrades
(asserts the full index is restored) and re-upgrades.
* docs(middlewares): tighten SandboxAudit and Clarification entries in AGENTS.md
PR #5134 grew agents/middlewares/AGENTS.md ~1.8 KB, pushing the effective
AGENTS.md chain for that directory over the 96 KiB hard limit once this
branch also documents scripts/benchmark/concurrency/ in backend/AGENTS.md.
Condense the two longest middleware entries (SandboxAuditMiddleware,
ClarificationMiddleware) without dropping any identifier, example, issue
reference, ordering constraint, or documented gap; chain back to ~96.8 KiB.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
BrowserSession scheduled three coroutines with a bare
asyncio.ensure_future(), so nothing held a reference to the resulting
tasks. The event loop only keeps weak references, so such a task can be
garbage collected before it finishes.
For the two live-frame schedulers the consequence is worse than losing
the task. Each sets a *_pending guard before scheduling and clears it in
a finally block:
self._settle_live_frames_pending = True
asyncio.ensure_future(self._settle_live_frames())
If the task is collected, the finally never runs, the guard stays True
forever, and every later _schedule_settle_live_frames()/
_schedule_input_live_frame() call returns early — silently stopping live
frame refresh for that session with no error.
Add _spawn_background(), which retains the task in a set and discards it
on completion, and route the three call sites through it. This matches
the pattern already used in task_tool, session_pool, notify and others.
Add regression tests asserting the task is retained across a gc.collect()
and released once it completes.
* fix(memory): cancel buffered extraction when agent is deleted or cleared
* fix(memory): cancel buffered work before agent delete
Address review: cancel before/after delete to close the rmtree race,
scope user_id=None cancels to the legacy root only, and import memory
helpers at module scope.
Signed-off-by: SPEC <zt1y17@soton.ac.uk>
* fix(memory): close remaining cancel races from review
Post-clear cancel, legacy-only all_agents scope, always cancel even when
memory is disabled, and fold cancel+delete into one offloaded thread.
Signed-off-by: SPEC <zt1y17@soton.ac.uk>
* docs(memory): align cancel_by_agent None-scope with legacy root
Document that user_id=None cancels only the legacy no-user bucket, matching
clear/storage semantics, not the whole process-local queue.
Signed-off-by: SPEC <zt1y17@soton.ac.uk>
* test(memory): fix cancel_by_agent docstring regression assertion
Signed-off-by: SPEC <zt1y17@soton.ac.uk>
* fix(memory): address final cancel review nits
Type the delete helper with AgentStore, replace docstring pinning with a
kwargs mapping test, and document scoped cancel + residual window in AGENTS.md.
Signed-off-by: SPEC <zt1y17@soton.ac.uk>
* fix(memory): resolve agent store inside delete worker thread
get_agent_store() does blocking config/FS work; keep it off the event
loop so test_delete_agent_does_not_block_event_loop and backend-blocking-io CI pass.
Signed-off-by: SPEC <zt1y17@soton.ac.uk>
---------
Signed-off-by: SPEC <zt1y17@soton.ac.uk>
* feat(auth): make login rate-limit parameters configurable, fixes#5108
Add auth.local.max_login_attempts (default 5) and auth.local.lockout_seconds
(default 300) so operators can tune the per-IP login throttle: raise the
ceiling for shared-egress-IP offices behind proxies/NAT, or tighten it for
stricter posture. Policy is live-read per call (matching the
_local_registration_enabled precedent), so a config reload applies without a
Gateway restart; raising the threshold mid-lockout immediately unblocks
affected IPs.
Review feedback addressed (willem-bd):
- Only FileNotFoundError falls back to the hardcoded defaults; a malformed
config propagates, mirroring _local_registration_enabled, so an operator
who tightened the policy never silently gets the more permissive defaults.
- _check_rate_limit looks up the record before resolving the policy, so a
clean IP pays zero config reads (get_app_config re-hashes config.yaml per
call and login_local is an unauthenticated async endpoint).
Bumps config_version to 39 in config.example.yaml and the Helm chart
(values.yaml + README example) so the chart drift check stays green.
* fix(auth): reject max_login_attempts=1 and honor live lockout_seconds for active lockouts
* fix(auth): close live-policy state gaps in login throttle (resurrection, count reset, broken-config verification)
* fix(auth): commit evaluated lockout duration on decreases too, preventing raise-resurrection
* test(auth): pin broken-config fail-closed sequence through the login route
* fix(auth): sweep expired locks by stored sentence and keep policy reads off the event loop
* fix(auth): re-read throttle record after the policy-resolution yield point
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
SSH_AUTH_SOCK points at the host's ssh-agent socket. A sandbox
subprocess that inherits it can sign and authenticate with every key
the agent holds (git push, ssh logins) without reading any key file --
the same credential-pointer leak class as the *_ASKPASS helpers the
env policy already scrubs deliberately. No wildcard pattern fits
(*AUTH* would strip benign names), so add an exact entry to
_BLOCKED_EXACT_NAMES.
A skill that genuinely needs the agent socket can still declare it via
required-secrets: injected values win over the blocklist by design.
Co-authored-by: zhouyujie <zhouyujie@keep.com>
