When a massive simultaneous disconnection occurs
(e.g. infra restart), all clients would reconnect
and invalidate their queries at exactly the same
time, causing a possible DB spike.
Adding random jitter spreads these events over a
time window so the load is absorbed gradually.
For the create_for_owner action, all the db operation are made in the
serializer. But the lock of the table was acquired in the viewsets, lot
of operation are made between the lock is made and the insert in db. We
move the lock operation closer to the insert in the database. We wrap it
in a transaction to release the lock once the commit made.
To test easily a build application with nginx,
we add a nginx-frontend to serve the static files
of the application, it will help us to test the
application in a more production-like environment.
The nginx conf was lacking the page reconciliation.
It is necessary to have it in place to avoid
404 errors when refreshing the page or accessing
a page directly.
It is a known issue when using the Next Router
in "export" mode, as it relies on client-side routing.
We want to be able to enable/disable the document
import feature for testing and gradual rollout
purposes. This commit adds a feature flag for
document import and updates the relevant components
and tests to respect this flag.
We want to control the conversion of document at upload time. We want to
disable this feature using a settings. The new settings
CONVERSION_UPLOAD_ENABLED should be used to enable or not the conversion
at upload feature. If disabled and a file is uploaded, the reponse will
return a 400
When a sub-sub-document had more than 20 children,
the pagination was not working.
This commit fixes the issue by ensuring that the
pagination logic is correctly applied to all
levels of the document tree.
I get
> TypeError: Cannot use 'in' operator to search for 'de' in undefined
when building MIT-only since `localesBNAI` is undefined then.
Signed-off-by: Maximilian Bosch <maximilian@mbosch.me>
By default, Node.js has a memory limit of
around 512MB, which can lead to out-of-memory
errors when processing large documents.
This commit increases the memory limit to
2GB for the y-provider server, allowing
it to handle larger documents without crashing.
We want to lock the table just before the insert we want to protect is
made. In the case of the perform_create action in the Document viewset,
an http call is made after the lock and can take a very long time,
blocking for nothing the table.
Add test_api_document_favorite_list_with_deleted_child to verify favorite_list
endpoint does not include deleted sub documents
Signed-off-by: Paul Vernin <paul.vernin@gmail.com>
The Yjs reader and writer in `convertHandler.ts`
were creating `Y.Doc`instances on every request
without calling `.destroy()`, causing a slow heap
leak that could crash the server.
Fixed by wrapping both sites in `try/finally`
blocks that call `ydoc.destroy()`.
Regression tests added to assert `destroy` is
called the expected number of times per request path.
We give a hint to the user about the minimum
number of characters required to perform a search
in the quick search input of the doc share modal.
This is to improve the user experience.
We upgraded vitest recently, we need to adapt
some of our tests to the new version.
We brought some modules improvments as well,
problemes that was highlighted by the new version
of vitest.
"eslint-plugin-import" is not well maintained anymore
better to use "eslint-plugin-import-x" which is a fork
of "eslint-plugin-import" and is actively maintained.
@react-pdf/renderer is not compatible with the
Blocknote version. We need to downgrade it to a
compatible version and pin it to avoid future issues.
When Blocknote updates to a compatible version,
we can upgrade @react-pdf/renderer again.
stylelint introduces lot of breaking changes
in its latest version, and since
we use it only for linting css files,
so we can block its upgrade for now and upgrade
it later when we will have more time to handle
the breaking changes.
Changed
- 💫(frontend) fix the help button to the bottom in tree #2073
- ♿️(frontend) improve version history list accessibility #2033
- ♿️(frontend) fix more options menu feedback for screen readers #2071
- ♿(frontend) focus skip link on headings and skip grid dropzone #1983
- ♿️(frontend) fix search modal accessibility issues #2054
- ♿️(frontend) add sr-only format to export download button #2088
- ♿️(frontend) announce formatting shortcuts for screen readers #2070
- ✨(frontend) add markdown copy icon for Copy as Markdown option #2096
- ♻️(backend) skip saving in database a document when payload is empty #2062
Fixed
- ♿️(frontend) fix aria-labels for table of contents #2065
- 🐛(backend) allow using search endpoint without refresh token enabled #2097
We add a debounce mechanism to the WebSocket
reconnect logic in the `useProviderStore` to
prevent rapid reconnection attempts that can
lead to performance issues and potential server
overload.
We observed a huge amount of logs sometimes in
the y-provider server logs, all related to the
same error: "No cookies".
When this happens, the client keeps trying to
reconnect, and the server keeps logging the error,
creating a loop.
We stop the loop by checking if the error is a
"No cookies" error, and if so, we don't
try to reconnect.
