fix: internal listeners infinite retry loop #284
Conversation
|
It would be good to cover this with tests |
src/PollingBlockTracker.ts
Outdated
| @@ -54,6 +52,10 @@ export class PollingBlockTracker | |||
|
|
|||
| private readonly _setSkipCacheFlag: boolean; | |||
|
|
|||
| readonly #onLatestBlockInternalListeners: (( | |||
There was a problem hiding this comment.
Other class variables use private readonly _, but I suppose we'll want to change them sometime in the future.
I'm ok with aligning this with others for now if that looks better
Co-authored-by: Mark Stacey <markjstacey@gmail.com>
This reverts commit eb5514d.
src/PollingBlockTracker.ts
Outdated
| const latestBlock: string = await new Promise((resolve) => | ||
| this.once('latest', resolve), | ||
| ); | ||
| const latestBlock: string = await new Promise((resolve) => { |
There was a problem hiding this comment.
To my understanding this change allows the PollingBlockTracker instance to be shut down (i.e. network calls will stop).
Though this promise here will never be rejected, meaning that in the case of:
PollingBlockTrackerbeing stopped, and- Desired network not available
The promise will be hanging forever.
I believe the fix should also reject this promise in the above scenario, resuming the caller execution
| this.#addInternalListener(onLatestBlockAvailable); | ||
| this.#addInternalListener(onLatestBlockUnavailable); | ||
| this.once('latest', onLatestBlockAvailable); | ||
| this.on('error', onLatestBlockUnavailable); |
There was a problem hiding this comment.
this listener is attached using .on instead of .once because the block tracker could be stopped after several errors, and we need to catch all of them
| ); | ||
| const latestBlock: string = await new Promise((resolve, reject) => { | ||
| // eslint-disable-next-line prefer-const | ||
| let onLatestBlockUnavailable: InternalListener; |
There was a problem hiding this comment.
This is first declared as uninitialized so that we can pass its reference to onLatestBlockAvailable. Unfortunately, it looks like eslint is unable to capture that
| if ( | ||
| this.listeners('error').filter((listener) => | ||
| this.#internalEventListeners.every( | ||
| (internalListener) => !Object.is(listener, internalListener), | ||
| ), | ||
| ).length === 0 | ||
| ) { | ||
| console.error(newErr); | ||
| } |
There was a problem hiding this comment.
The behavior of this try / catch block is unclear to me. It looks like we want to log the error when this.emit('error', newErr); throws, but looking at unit tests what we want is log the error when there is no listener attached to the error event.
Since we now have an internal listener on error as well, we need to exclude that
| this.#addInternalListener(onLatestBlockAvailable); | ||
| this.#addInternalListener(onLatestBlockUnavailable); | ||
| this.once('latest', onLatestBlockAvailable); | ||
| this.on('error', onLatestBlockUnavailable); |
There was a problem hiding this comment.
Hmm, I'm thinking that this would not cover the scenario where the block tracker is destroyed before fetching the first block and without throwing errors.
We can probably listen to _ended instead
There was a problem hiding this comment.
It looks like _ended could present other issues, because of the weird order of events received:
_startis emitted- Latest block is fetched from the provider
_endedevent listener is executedlatestevent listener is executed
Our block tracker implementation relies on the presence of listeners to establish whether the polling should be continued or stopped.
However, an internal listener is being added in the
getLatestBlockmethod, which will count as the other external listeners and will prevent the instance from stopping fetching new blocks. This creates an infinite loop in case the network is unreachable, because of the retry mechanism.This PR aims to fix this behavior by keeping track of internal listeners' references in order to exclude them from the listener's count
Fixes #163