Connection lifecycle work #2063
Conversation
* bump registry to v12.4.0 * use correct version * changeset
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
* Make tabs extensible * Add changeset * Remove console log
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
* Fix tabProps type * Add changset
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
I ran `shellcheck` on the script used to export a bundled minifront, suitable for inclusion in `pd`. These are small changes intended to make the script more resilient, e.g. failing fast and handling directories with spaces in the name. Prepared while working on penumbra-zone/penumbra#5089.
* fix(ui): #2053: add `smallTechnical` variant to Typography component * feat(ui): #2053: Add base for the ActionView component * feat(ui): #2053: Implement Spend and Output action views * feat(ui): #2053: implement SwapAction view * feat(ui): #2053: implement opaque views for existing actions * feat(ui): #2053: create skeleton files for unimplemented actions * chore: changeset * fix(ui): #2053: after review
turbocrime
added
testing
🦋 Changeset detectedLatest commit: 9770232 The changes in this PR will be included in the next version bump. This PR includes changesets to release 1 package
Not sure what this means? Click here to learn what changesets are. Click here if you're a maintainer who wants to add another changeset to this PR |
- manager is prepared for validation callback parameter - typed senders make it harder to emit a malformed message - incoming message checking is more thorough - poisons transport in fewer cases - matches errors up with relevant requests in more cases - fewer data clone failures
session-client will now automatically request to reconnect when disconnected. to support this, session-manager now uses a callback parameter to validate connection requests, instead of depending on external management of port injection.
* view service: LQT scaffolding (#2035) * view service: lqt voting notes scaffolding * view service: tournament votes rpc wip * wasm: extending planner with lqt voting action (#2037) * indexedDB: LQT historical votes table (#2038) * indexedDB: lqt historical votes table * workaround to normalize action cases * temporary measure for compilation purposes * linting * continue flushing out LQT integration * wasm: direct all available voting power to a single asset * wasm: informative expect messages * changeset * wasm: wasm bindgen tests * view server: expand vitest suite for new view service methods * wasm: clippy * protobuf: revert temporary workaround and consume fixed protos * action view: liquidity tournament visible and opaque views and translators * address feedback * linting * nit naming * refactor: account for LQT votes per delegation token * action views: fix opaque action view * more feedback * attempt to pass ci * satisfy linter * changeset * delete outdated changeset * stale ref; satisfy linter
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
There was a problem hiding this comment.
reviewed and left feedback on individual PRs that comprise this changeset
merged main into this PR and prax-wallet/prax#297 (which now includes prax-wallet/prax#296). I was still able to exploit the stream unreliability we discussed offline (albeit less often and it's improved after #2066) by cutting the connection during sync and waiting for the transport timeout before reloading the page – take note of the infinitely growing error logs (until page breaks). This feels more brittle compared to the previous behavior where streams could stay open indefinitely as long as the underlying chrome port wasn’t disconnected.
additionally, some stream-related tests were removed in #2066, so we should expand our stream test coverage.
Untitled-2.mov
and if you leave it running in that state, it eventually breaks the dapp with the following logs
and at that point, no amount of page reloads will resume syncing in the extension, and it remains in a zombie state.
after refreshing the extension in chrome://extensions/ while in that state after a while, it also randomly started resyncing the chain without me explicitly clearing the cache
| constructor(private incoming: chrome.runtime.Port) {} | ||
| constructor( | ||
| private incoming: chrome.runtime.Port, | ||
| private timeoutMs = 20_000, |
There was a problem hiding this comment.
question: this 20s idle stream timeout seems arbitrary, what about long-lived connections where the timeout may unintentionally break those connections? does increasing this lead to more frequent unreliability effects?
| if (this.timeoutMs) { | ||
| clearTimeout(this.timeout); | ||
| this.timeout = setTimeout( | ||
| () => cont.error(ConnectError.from('Source timeout', Code.DeadlineExceeded)), |
There was a problem hiding this comment.
question: what will the dev logs display when emitting DeadlineExceeded after stream timeout?
Description of Changes
Staging branch for improving reliability of connection lifecycle.
CRSessionClient, remove singleton pattern #2018 - second PR, corrects specific failure ofCRSessionClientblocking following work. no external api changesRelated Issue
CRSessionClientattempts to retain reference to a transferredMessagePortobject #1736