Revert per-resource wrap rethrow; record and continue

The earlier 'throw $e' aborted the batch on the first failure, leaving
items #6-N at STATUS_PENDING and skipping the rest of the source's
fetch loop. Restore the original 'absorb-and-continue' shape so every
resource in the batch is attempted and recorded — failures all land
in the migration report and the user retries the run cleanly.

Trade-off: library bugs (TypeError, etc.) caught here are absorbed and
don't reach Sentry. Acceptable — the deterministic-bug case fails the
same way for every resource, so absorbing produces a uniform 'all
errored' report instead of '1 errored + N pending'. User errors are
still kept out of Sentry by the worker's instanceof MigrationException
gate, which is the primary goal of this PR.
This commit is contained in:
Prem Palanisamy committed 2026-05-05 16:50:58 +01:00
1 parent e325782f76
commit da8205e8f3
1 file changed
+1 -3
+1 -3
View File
@@ -400,8 +400,6 @@ class Appwrite extends Destination
default => throw new \Exception('Invalid resource group', Exception::CODE_VALIDATION),
};
} catch (\Throwable $e) {
// Safety net for unexpected failures — record on the migration and
// rethrow so the worker boundary can route bugs to Sentry.
$resource->setStatus(Resource::STATUS_ERROR, $e->getMessage());
$this->addError(new Exception(
@@ -413,7 +411,7 @@ class Appwrite extends Destination
previous: $e
));
throw $e;
$responseResource = $resource;
} finally {
$this->dbForProject->setPreserveDates(false);
}