Repository navigation
Conversation
|
I reproduced the serial/parallel difference described in the PR using a Linux Java 11 backend with
Three failures occurred during checkpoint creation (HTTP 500), with corresponding server logs showing a
|
| }; | ||
|
|
||
| export const NotebookReadSkeleton = ({ core, onError, ...props }: NotebookReadSkeletonProps) => ( | ||
| <ReactErrorBoundary onError={onError}> |
There was a problem hiding this comment.
I confirmed that the view does not recover when navigating to another note after a post-mount error.
With note A displaying normally, I injected an error using the existing test-only failure prop, then cleared that prop. Navigating to note B within the same entry successfully fetched B and brought the host to ready, but the view remained empty. Leaving for the home page and reopening B restored the view.
Since this boundary receives no resetKey, it retains the previous error state across note changes. Even after the error condition is removed, opening another note does not recover the view; a fresh entry is required.
There was a problem hiding this comment.
Thanks for verifying this. I agree that navigating to another note after the error condition is cleared should recover the view without requiring a fresh entry. @xhaktm00 Could you address this and add a regression test in this PR?
voidmatcha
left a comment
There was a problem hiding this comment.
The new revision tests are skipped in the default CI run and aren’t included in the existing GitNotebookRepo step. Could we include them there so CI can catch regressions in revision loading and transitions back to the live note?
I’ve put together a suggested change that runs these cases with one worker in that step only, to avoid Git lock contention.
https://github.com/voidmatcha/zeppelin/tree/fix/ZEPPELIN-6727-revision-ci
This is a CI coverage suggestion. I haven’t found a current functional defect.
|
I missed @miinhho’s earlier comments and review when I left my review. Sorry about that. As miinhho confirmed, running with one worker avoids Git lock contention between the tests, so my suggestion was to run only the revision CI step sequentially in this PR. I also verified the suggested change against a real Zeppelin server with GitNotebookRepo. With one worker, retries disabled, and three repetitions across Chromium, Firefox, and WebKit, both authenticated and anonymous runs passed all 37 tests each. CI results I’ve opened ZEPPELIN-6769 to investigate and address the underlying server-side concurrent-write issue separately. |
NotebookReadPort is a NotebookPort whose snapshot is NotebookReadSnapshot. The private read host now declares its port with it, so a remote can read the read status and note without a cast while it keeps the same port object.
The note and revision routes have separate route configs, so moving between a note and a revision creates a new entry, while moving between notes, or between revisions, keeps it. The entry E2E now checks that moving between notes keeps the entry and that moving to and from a revision replaces it. The README says which moves let a failed React view render again.
9a3fa3c to
e3d1dca
Compare
|
@voidmatcha Thanks for the commits. I've included both, so the Git storage step now runs the revision tests with one worker. They pass locally on Git storage, both anonymous and authenticated. I also rebased onto master to resolve a conflict with ZEPPELIN-6703. @miinhho Thanks for catching this. Fixed in 2adf64f: after an error, the view renders again when another note opens, and there is a regression test for it. While checking this I found that moving between a note and a revision creates a new entry. The entry E2E now checks which moves keep the entry (e3d1dca), and the README says the same. |
What is this PR for?
Mount the React notebook read skeleton on the existing notebook URL, with an entry-time fallback to the ZEPPELIN-6737 Angular read view on the same host.
?notebookCoreRead¬ebookCoreReact, also under/revision/<revisionId>notebookCoreReact: the private route shows the ZEPPELIN-6737 view as beforenotebookCoreRead: the editable notebook route is unchanged?reactpath is not publishedTyped read port
NotebookReadPort(zeppelin-notebook-core): aNotebookPortwhose snapshot isNotebookReadSnapshotNotebookCoreReadHost.portnow has this type. It is the same frozen object and still fits wherever aNotebookPortis expectedNotebookReadProviderpasses that exact object toNotebookCoreProvider(ZEPPELIN-6726) and exposes it touseNotebookReadSnapshot(), so the remote reads the read state without a castReact skeleton (
./NotebookReadSkeleton)onMountedafter its first commit, and again after it recovers from an error. This means the mount completed, not that the notebook is readyReactErrorBoundaryandonErrorfailure: 'mount' | 'after-mount'throws during the first render, or on the first render after the mount completedEntry and fallback (
NotebookCoreEntryComponent)canMatchguard (ZEPPELIN-6737)NotebookCoreReadHostfor the route lifetime: opens the note or revision from the route parameters and disposes the host onceNotebookCoreReadComponentnow receives that host as a required input and no longer opens or disposes itReactMountDirectivewith the shared loader and timeout), never bothparagraphquery parameter, the failure value,onMountedandonError. The React view does not read the URL itselfdata-rendererangularnotebookCoreReact: the Angular read viewreact-loadingreact-mountedangular-fallbackruntime-errorruntime-errorleaves the view empty on purpose: the issue rules out an automatic switch after the mountparagraphin the URL but does not scroll to it, because the ZEPPELIN-6737 view has no anchor handlingnotebookCoreReaddestroys the entry, which disposes the host and its subscriptions, and the editable notebook loads the note itself. This is not the same-port fallbackWhat type of PR is it?
Feature
Todos
NotebookReadPortand read it in React throughNotebookReadProvideranduseNotebookReadSnapshot./NotebookReadSkeletonremote with a mount report and test-only failure injectionnotebookCoreReactflagWhat is the Jira issue?
https://issues.apache.org/jira/browse/ZEPPELIN-6727
How should this be tested?
From
zeppelin-web-angular:Real-route Playwright (
e2e/tests/notebook/core-read/notebook-core-entry.spec.ts, 13 tests; Chromium, one worker, local server with Git notebook storage):conf/shiro.ini.templatewithadminadded touser1, as in CI): 13 passednotebook-core-read.spec.ts: 7 passed and 1 skipped anonymous, 8 passed authenticatedremoteEntry.jsto an abort, a 404 or a request that never answers. Each fallback test waits for that request, so the Angular view is a fallback rather than the no-flag viewNot run locally:
read-fixtures.spec.ts(@live, needs a capture server)Revision tests on CI:
VFSNotebookRepo). The Git storage step now runs them with one workernotebook/.git/indexlock. This is tracked in ZEPPELIN-6769Screenshots (if appropriate)
Questions: