map unexpected fetch AbortError to ConnectionError, so doesn't stop sync

This commit is contained in:
Bruno Windels 2021-04-09 14:25:19 +02:00
parent 2341748023
commit 2b1f4866a9

View file

@ -47,7 +47,23 @@ class RequestWrapper {
}, err => {
// if this._requestResult is still set, the abort error came not from calling abort here
if (err.name === "AbortError" && this._requestResult) {
const err = new Error(`Unexpectedly aborted, see #187.`);
// The service worker sometimes (only on Firefox, on long, large request,
// perhaps it has its own timeout?) aborts the request, see #187.
// When it happens, the best thing to do seems to be to retry.
//
// In the service worker, we will also actively abort requests when trying to
// get a new service worker to activate, as the service worker will only be replaced
// when there are no more (fetch) events for the current one to handle.
//
// In that case, the request function (in fetch.js) will check
// the haltRequests flag on the service worker handler, and it will
// actually not do any requests, as that would break the update process.
//
// So it is OK to return a timeout ConnectionError here.
// If we're updating the service worker, the /versions polling will
// actually be blocked at the fetch level because haltRequests is set.
// And for #187, retrying is the right thing to do.
const err = new ConnectionError(`Service worker aborted, either updating or hit #187.`, true);
log?.catch(err);
throw err;
} else {