feat: cascade TTL to container sub-tasks (rapyer >=1.3.4) - #137
feat: cascade TTL to container sub-tasks (rapyer >=1.3.4)#137yedidyakfir wants to merge 14 commits into
Conversation
Bump the rapyer cap to >=1.3.4,<1.4.0 across third-magic, mageflow-mcp and mage-voyance and re-lock every workspace. Adapt to 1.3.x changes: - instance apipeline now re-fetches on entry, so add_tasks uses ignore_redis_error=True (the swarm is not yet committed during the atomic create + add_tasks pipeline). - RedisConfig is now a pydantic model, so apply_ttl_config and the TTL test helpers use model_copy(update=...) instead of dataclasses.replace. Full unit suites green on 1.3.4 (third-magic 134, mageflow 263). Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Declare a sub_task_refs ForeignKey collection on SwarmTaskSignature and ChainTaskSignature, annotated with CascadeTTL(), mirroring the child tasks. Containers get a Meta with refresh_ttl scoped to write actions, so any write refreshes the container's TTL and cascades that refresh to its sub-task edges (real-Redis cascade; fakeredis re-arms root only). The existing tasks key-lists and their consumers are untouched — the FK field is an additional cascade edge, populated at chain creation and in swarm add_tasks. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Cascade edge-following needs Redis Functions + RedisJSON, which fakeredis cannot emulate, so add an integration suite that runs against a real Redis Stack (THIRDMAGIC_TEST_REDIS_URL, default redis://localhost:6379) and skips when unavailable or the RedisJSON module is missing. Verifies that a write to a swarm/chain extends its sub-tasks' TTL. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
Important Review skippedAuto incremental reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Plus Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 WalkthroughWalkthroughChangesContainer TTL Cascade
MCP Documentation Server
Estimated code review effort: 3 (Moderate) | ~20 minutes Sequence Diagram(s)sequenceDiagram
participant Container as Chain/Swarm Container
participant Redis
participant Task as Referenced TaskSignature
Container->>Redis: Perform configured write action
Redis->>Task: Cascade TTL refresh through sub_task_refs
Possibly related PRs
🚥 Pre-merge checks | ✅ 2 | ❌ 3❌ Failed checks (3 warnings)
✅ Passed checks (2 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 5
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In @.codex/config.toml:
- Around line 2-5: Pin the mcp-remote dependency at an approved exact version in
the frontend package manifest and lockfile, then update the command
configuration to invoke that installed pinned package rather than the
unversioned npx resolution. Preserve the existing MCP endpoint and arguments
while changing the executable reference.
In `@libs/third-magic/tests/integration/conftest.py`:
- Around line 45-49: Update the mock_adapter fixture to snapshot
Signature.ClientAdapter before replacing it, then restore the original value in
a finally block surrounding the yield. Keep yielding the MagicMock adapter while
the test runs and ensure restoration occurs even when the test raises.
- Around line 26-42: Update the real_redis fixture to require a dedicated test
Redis endpoint/database before running integration tests, and replace both
client.flushall() calls with flushdb(). Preserve the existing setup and teardown
behavior while ensuring cleanup only affects the selected database, not every
database on the configured Redis server.
In `@libs/third-magic/tests/integration/test_cascade_ttl.py`:
- Around line 29-31: Expand both cascade TTL test sites in
libs/third-magic/tests/integration/test_cascade_ttl.py#L29-L31 and `#L48-L50` with
distinct parent and child TTL configurations. Assert the parent key refreshes
near its configured TTL, while every child key is re-armed near its own
configured TTL rather than the container’s value.
In `@libs/third-magic/tests/unit/test_cascade_ttl.py`:
- Line 20: Update the pytest.mark.parametrize declaration for container_cls to
use the project-required list-of-lists case format, while preserving the
existing SwarmTaskSignature and ChainTaskSignature parameter values.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro Plus
Run ID: 6f715760-2cb0-495f-bc21-447e7e0d9870
⛔ Files ignored due to path filters (9)
.DS_Storeis excluded by!**/.DS_Storelibs/.DS_Storeis excluded by!**/.DS_Storelibs/mage-voyance/uv.lockis excluded by!**/*.lock,!**/*.lock,!**/uv.locklibs/mageflow-e2e/uv.lockis excluded by!**/*.lock,!**/*.lock,!**/uv.locklibs/mageflow-mcp/uv.lockis excluded by!**/*.lock,!**/*.lock,!**/uv.locklibs/mageflow/.DS_Storeis excluded by!**/.DS_Storelibs/mageflow/uv.lockis excluded by!**/*.lock,!**/*.lock,!**/uv.locklibs/third-magic/.DS_Storeis excluded by!**/.DS_Storelibs/third-magic/uv.lockis excluded by!**/*.lock,!**/*.lock,!**/uv.lock
📒 Files selected for processing (16)
.codex/config.tomlCHANGELOG.mdlibs/mage-voyance/pyproject.tomllibs/mageflow-mcp/pyproject.tomllibs/mageflow/mageflow/config.pylibs/mageflow/tests/unit/test_remove_ttl.pylibs/mageflow/tests/unit/test_ttl_config.pylibs/third-magic/pyproject.tomllibs/third-magic/tests/integration/__init__.pylibs/third-magic/tests/integration/conftest.pylibs/third-magic/tests/integration/test_cascade_ttl.pylibs/third-magic/tests/unit/test_cascade_ttl.pylibs/third-magic/thirdmagic/chain/creator.pylibs/third-magic/thirdmagic/chain/model.pylibs/third-magic/thirdmagic/container.pylibs/third-magic/thirdmagic/swarm/model.py
| command = "npx" | ||
| args = [ | ||
| "mcp-remote", | ||
| "https://gitmcp.io/imaginary-cherry/rapyer", |
There was a problem hiding this comment.
🔒 Security & Privacy | 🟠 Major | ⚡ Quick win
🧩 Analysis chain
🏁 Script executed:
#!/bin/sh
set -eu
echo "== files =="
git ls-files | grep -E '(^|/)\.codex/config\.toml|\.mcp\.json|package(-lock)?\.json|pnpm-lock\.yaml|yarn\.lock|package-lock\.json' || true
echo
echo "== .codex/config.toml =="
cat -n .codex/config.toml 2>/dev/null || true
echo
echo "== search mcp-remote usage =="
rg -n "mcp-remote|mcp" .codex .mcp.json package.json package-lock.json pnpm-lock.yaml yarn.lock 2>/dev/null || trueRepository: imaginary-cherry/mageflow
Length of output: 817
🌐 Web query:
npx mcp-remote executable package version unpinned security supply chain documentation
💡 Result:
The use of npx to execute mcp-remote introduces significant supply chain security risks if package versions are not pinned. The practice of using unpinned commands like npx mcp-remote (or npx -y mcp-remote) allows the package manager to fetch the latest available version from the npm registry, which can lead to the silent introduction of malicious code or vulnerable versions [1][2][3]. A critical security vulnerability, CVE-2025-6514, was disclosed in July 2025 affecting mcp-remote versions 0.0.5 through 0.1.15 [4][2]. This vulnerability enabled remote code execution (RCE) via OS command injection when a user connected to an untrusted MCP server that provided a malicious authorization_endpoint URL [4][5]. This flaw highlights the dangers of running unpinned, automated tooling in environments with access to sensitive local credentials and system resources [1][6]. To secure your MCP environment, implement the following supply chain hygiene practices: 1. Pin Exact Versions: Never use "latest" or semver ranges (e.g., ^0.1.16) for MCP servers or proxies. Pin to an exact version in your configuration and lockfiles (e.g., "mcp-remote": "0.1.16") [1][2][3]. 2. Avoid Automated "npx" Patterns: In production or sensitive development environments, replace npx one-liners with explicit local installations. Use npm install to manage versions via a package.json and package-lock.json or equivalent, which allows for integrity hashing and consistent dependency resolution [7][2][3]. 3. Audit Dependencies: Treat every MCP server as a third-party dependency. Before installation, review the source code, audit the manifest for requested permissions (e.g., filesystem access, shell commands), and verify the dependency tree for known vulnerabilities [3][8]. 4. Principle of Least Privilege: If you must use a proxy like mcp-remote, ensure it is updated to at least version 0.1.16 [4][1]. Run MCP servers in isolated environments (such as containers or restricted OS users) with no ambient cloud credentials and limited filesystem access [9][6]. 5. Enforce Authentication and Transport Security: Connect only to trusted servers over HTTPS and require secure authentication standards, such as OAuth 2.1 with PKCE, to mitigate intercept and man-in-the-middle attacks [1][2][9].
Citations:
- 1: https://teqvolt.com/tech-tips/mcp-security-2026-hardening-playbook
- 2: https://policyascode.dev/blog/mcp-security-vulnerabilities-2026/
- 3: https://gethasp.com/guides/mcp-supply-chain-hygiene/
- 4: https://jfrog.com/blog/2025-6514-critical-mcp-remote-rce-vulnerability/
- 5: https://research.jfrog.com/vulnerabilities/mcp-remote-command-injection-rce-jfsa-2025-001290844/
- 6: https://www.exploreagentic.ai/insights/mcp-server-security-hardening/
- 7: https://developertoolkit.ai/en/shared-workflows/mcp-ecosystem/mcp-security/
- 8: https://drel.ai/blog/mcp-server-supply-chain
- 9: https://safeguard.sh/resources/blog/securing-mcp-servers-a-practical-checklist
Pin the mcp-remote executable version.
npx mcp-remote resolves the latest package version from npm at each invocation, so this agent MCP tool can execute unreviewed code or regress to a vulnerable version. Add mcp-remote to the frontend package/lockfile at an approved exact version and reference the pinned package in .codex/config.toml instead of the unversioned uninstalled command.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In @.codex/config.toml around lines 2 - 5, Pin the mcp-remote dependency at an
approved exact version in the frontend package manifest and lockfile, then
update the command configuration to invoke that installed pinned package rather
than the unversioned npx resolution. Preserve the existing MCP endpoint and
arguments while changing the executable reference.
| @pytest_asyncio.fixture | ||
| async def real_redis(): | ||
| client = aioredis.Redis.from_url(REDIS_URL) | ||
| try: | ||
| await client.ping() | ||
| except Exception: | ||
| pytest.skip(f"No Redis reachable at {REDIS_URL}") | ||
| if not await _has_json_module(client): | ||
| await client.aclose() | ||
| pytest.skip(f"Redis at {REDIS_URL} lacks the RedisJSON module") | ||
| await client.flushall() | ||
| await rapyer.init_rapyer(client) | ||
| try: | ||
| yield client | ||
| finally: | ||
| await client.flushall() | ||
| await client.aclose() |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win
Avoid FLUSHALL on a configurable Redis endpoint.
This erases every database on whatever server THIRDMAGIC_TEST_REDIS_URL targets. Require a dedicated test endpoint/database and use FLUSHDB instead.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@libs/third-magic/tests/integration/conftest.py` around lines 26 - 42, Update
the real_redis fixture to require a dedicated test Redis endpoint/database
before running integration tests, and replace both client.flushall() calls with
flushdb(). Preserve the existing setup and teardown behavior while ensuring
cleanup only affects the selected database, not every database on the configured
Redis server.
| @pytest.fixture | ||
| def mock_adapter(): | ||
| adapter = MagicMock(spec=BaseClientAdapter) | ||
| Signature.ClientAdapter = adapter | ||
| yield adapter |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Restore the global client adapter after the fixture.
Signature.ClientAdapter remains a MagicMock after teardown, contaminating later tests. Snapshot it before assignment and restore it in a finally block.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@libs/third-magic/tests/integration/conftest.py` around lines 45 - 49, Update
the mock_adapter fixture to snapshot Signature.ClientAdapter before replacing
it, then restore the original value in a finally block surrounding the yield.
Keep yielding the MagicMock adapter while the test runs and ensure restoration
occurs even when the test raises.
| # Assert | ||
| for task in tasks: | ||
| assert await real_redis.ttl(task.key) > SHORT_TTL |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Verify that cascading uses each child’s TTL, not the container TTL.
Both assertions pass if children are incorrectly refreshed with the parent’s TTL. Configure distinct parent/child TTLs, assert the parent refreshes, and assert each child is re-armed near its own configured value.
libs/third-magic/tests/integration/test_cascade_ttl.py#L29-L31: add distinct swarm/task TTL assertions.libs/third-magic/tests/integration/test_cascade_ttl.py#L48-L50: add distinct chain/task TTL assertions.
As per path instructions, **/tests/**: “Focus on test coverage completeness and edge cases.”
📍 Affects 1 file
libs/third-magic/tests/integration/test_cascade_ttl.py#L29-L31(this comment)libs/third-magic/tests/integration/test_cascade_ttl.py#L48-L50
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@libs/third-magic/tests/integration/test_cascade_ttl.py` around lines 29 - 31,
Expand both cascade TTL test sites in
libs/third-magic/tests/integration/test_cascade_ttl.py#L29-L31 and `#L48-L50` with
distinct parent and child TTL configurations. Assert the parent key refreshes
near its configured TTL, while every child key is re-armed near its own
configured TTL rather than the container’s value.
Source: Path instructions
| ) | ||
|
|
||
|
|
||
| @pytest.mark.parametrize("container_cls", [SwarmTaskSignature, ChainTaskSignature]) |
There was a problem hiding this comment.
📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win
Use list-of-lists for parametrization.
Change the parameter cases to the project-required list-of-lists format.
Proposed fix
-@pytest.mark.parametrize("container_cls", [SwarmTaskSignature, ChainTaskSignature])
+@pytest.mark.parametrize(
+ "container_cls",
+ [[SwarmTaskSignature], [ChainTaskSignature]],
+)As per path instructions, pytest.mark.parametrize must use list-of-lists format.
📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| @pytest.mark.parametrize("container_cls", [SwarmTaskSignature, ChainTaskSignature]) | |
| `@pytest.mark.parametrize`( | |
| "container_cls", | |
| [[SwarmTaskSignature], [ChainTaskSignature]], | |
| ) |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@libs/third-magic/tests/unit/test_cascade_ttl.py` at line 20, Update the
pytest.mark.parametrize declaration for container_cls to use the
project-required list-of-lists case format, while preserving the existing
SwarmTaskSignature and ChainTaskSignature parameter values.
Source: Path instructions
Match the existing mageflow integration convention: the real_redis fixture connects to THIRDMAGIC_TEST_REDIS_URL and lets a missing/ unsuitable Redis surface as an error, rather than silently skipping. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Address review: don't add a parallel sub_task_refs field. Convert the existing chain/swarm tasks field itself to Annotated[list[Reference[TaskSignature]], CascadeTTL()] so it is both the child list and the cascade edge. task_ids returns the target keys, and internal/test consumers that treated tasks as keys now use task_ids. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
task_ids builds the full key list; chain.acall and on_sub_task_done only need one element, so read the target_key directly off self.tasks instead of indexing into the materialized list. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
|
|
||
| async def sub_tasks(self) -> list[TaskSignature]: | ||
| sub_tasks = await rapyer.afind(*self.tasks, skip_missing=True) | ||
| sub_tasks = await rapyer.afind(*self.task_ids, skip_missing=True) |
There was a problem hiding this comment.
Can't you use the reference interface for this? It will be more efficient, no?
|
|
||
| async def acall(self, msg: Any, set_return_field: bool = True, **kwargs): | ||
| first_task = await rapyer.afind_one(self.tasks[0]) | ||
| first_task = await rapyer.afind_one(self.task_ids[0]) |
There was a problem hiding this comment.
If you take id from only on task, don't call tasks IDs which iterate over the entire list
| async def suspend(self): | ||
| await asyncio.gather( | ||
| *[TaskSignature.suspend_from_key(task_id) for task_id in self.tasks], | ||
| *[TaskSignature.suspend_from_key(task_id) for task_id in self.task_ids], |
There was a problem hiding this comment.
You can iterate over the tasks and get the tasks id from each one instead of 2 iterations
| async def interrupt(self): | ||
| await asyncio.gather( | ||
| *[TaskSignature.interrupt_from_key(task_id) for task_id in self.tasks], | ||
| *[TaskSignature.interrupt_from_key(task_id) for task_id in self.task_ids], |
There was a problem hiding this comment.
You can iterate over the tasks and get the tasks id from each one instead of 2 iterations
| async def resume(self): | ||
| await asyncio.gather( | ||
| *[TaskSignature.resume_from_key(task_key) for task_key in self.tasks], | ||
| *[TaskSignature.resume_from_key(task_key) for task_key in self.task_ids], |
There was a problem hiding this comment.
You can iterate over the tasks and get the tasks id from each one instead of 2 iterations
| CONTAINER_WRITE_ACTIONS = ( | ||
| ActionGroup.CREATE | ||
| | ActionGroup.UPDATE | ||
| | ActionGroup.APPEND | ||
| | ActionGroup.ERASE | ||
| | ActionGroup.ARITHMETIC |
There was a problem hiding this comment.
Update and create is enough, the rest are sub categories
|
|
||
| async def sub_tasks(self) -> list[TaskSignature]: | ||
| tasks = await rapyer.afind(*self.tasks) | ||
| tasks = await rapyer.afind(*self.task_ids) |
There was a problem hiding this comment.
Again, use the reference API instea
| async def change_status(self, status: SignatureStatus): | ||
| paused_chain_tasks = [ | ||
| TaskSignature.safe_change_status(task, status) for task in self.tasks | ||
| TaskSignature.safe_change_status(task, status) for task in self.task_ids |
There was a problem hiding this comment.
You can iterate over the tasks and get the tasks id from each one instead of 2 iterations
| async def suspend(self): | ||
| await asyncio.gather( | ||
| *[TaskSignature.suspend_from_key(swarm_id) for swarm_id in self.tasks], | ||
| *[TaskSignature.suspend_from_key(swarm_id) for swarm_id in self.task_ids], |
There was a problem hiding this comment.
You can iterate over the tasks and get the tasks id from each one instead of 2 iterations
| async def resume(self): | ||
| await asyncio.gather( | ||
| *[TaskSignature.resume_from_key(task_id) for task_id in self.tasks], | ||
| *[TaskSignature.resume_from_key(task_id) for task_id in self.task_ids], |
There was a problem hiding this comment.
You can iterate over the tasks and get the tasks id from each one instead of 2 iterations
…efresh actions - sub_tasks now resolves via the Reference API (ref.afetch) instead of rebuilding keys and calling afind. - Container fan-outs (change_status/suspend/interrupt/resume) iterate self.tasks once and read ref.target_key, instead of iterating the materialized task_ids list. - refresh_ttl is just CREATE | UPDATE (APPEND/ERASE/ARITHMETIC folded in). - chain acall / on_sub_task_done already read a single ref directly. - astatus test loads the container first, matching how mageflow.astatus invokes it (fresh, unresolved references). Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
| self.tasks.extend(task_keys) | ||
| self.tasks.extend([Reference(task) for task in tasks]) | ||
| self.tasks_left_to_run.extend(task_keys) | ||
| await self.asave() |
There was a problem hiding this comment.
This is very bad, why do we need to save after apipeline? it saves the changes automaticaly, asave can override changes from parallel worker and create race condition
- on_sub_task_done reads sub_task.chain_index (stored on each chain member at creation) and indexes the next task directly, instead of scanning the whole tasks list. chain_index lives on Signature so chain members of any type (task/swarm/chain) carry it. - sub_tasks goes back to a single batched rapyer.afind (one Redis round trip) over the references' target keys, rather than per-reference afetch. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Replace the leaky chain_index-on-Signature with a current_index pointer on ChainTaskSignature. on_sub_task_done uses it as an O(1) fast path, and only falls back to a scan (re-syncing the pointer) when the pointer does not match the completed task — which also keeps it correct under retry double-fires. Tasks no longer carry any chain-position state. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… asave in add_tasks - add_tasks no longer calls asave(); reassigning self.tasks inside the apipeline persists safely without a blind full-document overwrite that could clobber a concurrent worker (review feedback). - Integration tests spin up a real Redis Stack via testcontainers (AsyncRedisContainer) instead of a fixed URL, session-scoped, and run in CI (new 'Run Integration Tests' step; ubuntu runners have Docker). - Add testcontainers[redis] to thirdmagic dev deps; ignore .DS_Store/.codex. Refs #136 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
rapyer.afind() with no keys returns every model in the DB. Several call sites spread a possibly-empty iterable into it (task callbacks, sub_tasks), so an empty set silently triggered a full-DB fetch — worst on every task completion via activate_success/activate_error, where callbacks are usually empty. Add afind_keys_guarded(keys) in thirdmagic.utils that short-circuits to [] for an empty key set, and route the unguarded sites through it. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ge235p3Le4aLDjmP1AubRT
The `tasks` field annotation is a list[Reference] but rapyer converts it to a RedisList at runtime, so `.extend()` is available and issues an atomic append. Replace the read-modify-write reassignment (self.tasks = self.tasks + [...]), which could clobber a concurrent worker's additions, with a race-safe extend. Verified: thirdmagic unit (139) and the swarm cascade integration test pass, so TTL cascade still fires on the extended list. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ge235p3Le4aLDjmP1AubRT
1 participant
Closes #136
Summary
Upgrade rapyer to
>=1.3.4and use its new features so that writing to a container signature (swarm / chain) refreshes its TTL and cascades that refresh to all of its sub-tasks, keeping the whole subtree alive together while the container is active. Purely declarative model config — no explicitupdate_ttlmethod.What changed
>=1.3.4,<1.4.0acrossthird-magic,mageflow-mcp,mage-voyance; re-locked every workspace. Adapted to 1.3.x breaking changes:apipelinere-fetches on entry → swarmadd_tasksusesignore_redis_error=True(model not yet committed during atomic create+add).RedisConfigis now a pydantic model →apply_ttl_config+ TTL test helpers usemodel_copyinstead ofdataclasses.replace.SwarmTaskSignature/ChainTaskSignaturegain asub_task_refs: Annotated[list[Reference[TaskSignature]], CascadeTTL()]field mirroring their children, plus aMetawithrefresh_ttlscoped to write actions. The existingtaskskey-lists and all their consumers are untouched.sub_task_refspersistence) + a real-Redis integration suite that verifies a container write extends its sub-tasks' TTL.Cascade semantics
rapyer re-arms each reached child to its own
Meta.ttl(keep-the-subtree-alive-together), not "copy one number onto every child". With mageflow's uniform active TTL this is the intended behavior.Testing note
Cascade edge-following needs a real Redis Stack (Functions + RedisJSON); fakeredis only re-arms the root. The integration suite (
libs/third-magic/tests/integration) runs againstTHIRDMAGIC_TEST_REDIS_URLand skips when no JSON-capable Redis is present. Verified locally againstredis-stack-server.🤖 Generated with Claude Code
Summary by CodeRabbit
New Features
Bug Fixes
Changelog