In a multi-main setup, we have the following issue. The user's client
connects to main A and runs a test webhook, so main A starts listening
for a webhook call. A third-party service sends a request to the test
webhook URL. The request is forwarded by the load balancer to main B,
who is not listening for this webhook call. Therefore, the webhook call
is unhandled.
To start addressing this, cache test webhook registrations, using Redis
for queue mode and in-memory for regular mode. When the third-party
service sends a request to the test webhook URL, the request is
forwarded by the load balancer to main B, who fetches test webhooks from
the cache and, if it finds a match, executes the test webhook. This
should be transparent - test webhook behavior should remain the same as
so far.
Notes:
- Test webhook timeouts are not cached. A timeout is only relevant to
the process it was created in, so another process retrieving from Redis
a "foreign" timeout will be unable to act on it. A timeout also has
circular references, so `cache-manager-ioredis-yet` is unable to
serialize it.
- In a single-main scenario, the timeout remains in the single process
and is cleared on test webhook expiration, successful execution, and
manual cancellation - all as usual.
- In a multi-main scenario, we will need to have the process who
received the webhook call send a message to the process who created the
webhook directing this originating process to clear the timeout. This
will likely be implemented via execution lifecycle hooks and Redis
channel messages checking session ID. This implementation is out of
scope for this PR and will come next.
- Additional data in test webhooks is not cached. From what I can tell,
additional data is not needed for test webhooks to be executed.
Additional data also has circular references, so
`cache-manager-ioredis-yet` is unable to serialize it.
Follow-up to: #8155
This PR simplifies state in test webhooks so that it can be cached
easily. Caching this state will allow us to start using Redis for manual
webhooks, to support manual webhooks to work in multi-main setup.
- [x] Convert `workflowWebhooks` to a getter - no need to optimize for
deactivation
- [x] Remove array from value in `TestWebhooks.webhookUrls`
- [x] Consolidate `webhookUrls` and `registeredWebhooks`
Remove duplication, improve readability, and expand tests for
`TestWebhooks.ts` - in anticipation for storing test webhooks in Redis.
---------
Co-authored-by: कारतोफ्फेलस्क्रिप्ट™ <aditya@netroy.in>
Followup to #7566 | Story: https://linear.app/n8n/issue/PAY-926
### Manual workflow activation and deactivation
In a multi-main scenario, if the user manually activates or deactivates
a workflow, the process (whether leader or follower) that handles the
PATCH request and updates its internal state should send a message into
the command channel, so that all other main processes update their
internal state accordingly:
- Add to `ActiveWorkflows` if activating
- Remove from `ActiveWorkflows` if deactivating
- Remove and re-add to `ActiveWorkflows` if the update did not change
activation status.
After updating their internal state, if activating or deactivating, the
recipient main processes should push a message to all connected
frontends so that these can update their stores and so reflect the value
in the UI.
### Workflow activation errors
On failure to activate a workflow, the main instance should record the
error in Redis - main instances should always pull activation errors
from Redis in a multi-main scenario.
### Leadership change
On leadership change...
- The old leader should stop pruning and the new leader should start
pruning.
- The old leader should remove trigger- and poller-based workflows and
the new leader should add them.
This change expands on the command channel communication introduced
lately between the main instance(s) and the workers. The frontend gets a
new menu entry "Workers" which will, when opened, trigger a regular call
to getStatus from the workers. The workers then respond via their
response channel to the backend, which then pushes the status to the
frontend.
This introduces the use of ChartJS for metrics.
This feature is still in MVP state and thus disabled by default for the
moment.
- Enable two-way communication with web sockets
- Enable sending push messages to specific users
- Add collaboration service for managing active users for workflow
Missing things:
- State is currently kept only in memory, making this not work in
multi-master setups
- Removing a user from active users in situations where they go inactive
or we miss the "workflow closed" message
- I think a timer based solution for this would cover most edge cases.
I.e. have FE ping every X minutes, BE removes the user unless they have
received a ping in Y minutes, where Y > X
- FE changes to be added later by @MiloradFilipovic
Github issue / Community forum post (link here to close automatically):
---------
Co-authored-by: कारतोफ्फेलस्क्रिप्ट™ <aditya@netroy.in>
Story: https://linear.app/n8n/issue/PAY-926
This PR coordinates workflow activation on instance startup and on
leadership change in multiple main scenario in the internal API. Part 3
on manual workflow activation and deactivation will be a separate PR.
### Part 1: Instance startup
In multi-main scenario, on starting an instance...
- [x] If the instance is the leader, it should add webhooks, triggers
and pollers.
- [x] If the instance is the follower, it should not add webhooks,
triggers or pollers.
- [x] Unit tests.
### Part 2: Leadership change
In multi-main scenario, if the main instance leader dies…
- [x] The new main instance leader must activate all trigger- and
poller-based workflows, excluding webhook-based workflows.
- [x] The old main instance leader must deactivate all trigger- and
poller-based workflows, excluding webhook-based workflows.
- [x] Unit tests.
To test, start two instances and check behavior on startup and
leadership change:
```
EXECUTIONS_MODE=queue N8N_LEADER_SELECTION_ENABLED=true N8N_LICENSE_TENANT_ID=... N8N_LICENSE_ACTIVATION_KEY=... N8N_LOG_LEVEL=debug npm run start
EXECUTIONS_MODE=queue N8N_LEADER_SELECTION_ENABLED=true N8N_LICENSE_TENANT_ID=... N8N_LICENSE_ACTIVATION_KEY=... N8N_LOG_LEVEL=debug N8N_PORT=5679 npm run start
```
Based on #7065 | Story: https://linear.app/n8n/issue/PAY-771
n8n on filesystem mode marks binary data to delete on manual execution
deletion, on unsaved execution completion, and on every execution
pruning cycle. We later prune binary data in a separate cycle via these
marker files, based on the configured TTL. In the context of introducing
an S3 client to manage binary data, the filesystem mode's mark-and-prune
setup is too tightly coupled to the general binary data management
client interface.
This PR...
- Ensures the deletion of an execution causes the deletion of any binary
data associated to it. This does away with the need for binary data TTL
and simplifies the filesystem mode's mark-and-prune setup.
- Refactors all execution deletions (including pruning) to cause soft
deletions, hard-deletes soft-deleted executions based on the existing
pruning config, and adjusts execution endpoints to filter out
soft-deleted executions. This reduces DB load, and keeps binary data
around long enough for users to access it when building workflows with
unsaved executions.
- Moves all execution pruning work from an execution lifecycle hook to
`execution.repository.ts`. This keeps related logic in a single place.
- Removes all marking logic from the binary data manager. This
simplifies the interface that the S3 client will meet.
- Adds basic sanity-check tests to pruning logic and execution deletion.
Out of scope:
- Improving existing pruning logic.
- Improving existing execution repository logic.
- Adjusting dir structure for filesystem mode.
---------
Co-authored-by: कारतोफ्फेलस्क्रिप्ट™ <aditya@netroy.in>
