Mentat

I think I know what the issue was and luckily, we didn't lose any events.

To reproduce this bug, try stopping the storage just after this line appears in the log:

2024-06-13 11:28:50,289 mentat-storage.py [5269] INFO: Component 'storage': Commit timeout '5' elapsed, performing commit.

That is produced by this function:

def cbk_event_database_commit(self, daemon, args):
    """ 
    Check, whether there are any messages waiting to be committed for greater than
    configured time period. Commit them if necessary to avoid loss of data.
    """ 
    daemon.logger.debug(
        "Component '{}': Checking whether commit needs to be enforced".format(
            self.cid
        )
    )
    if self.events_uncommitted and ((time.time() - self.last_commit) > self.commit_bulkintv):
        daemon.logger.info(
            "Component '{}': Commit timeout '{}' elapsed, performing commit.".format(
                self.cid,
                self.commit_bulkintv
            )
        )
        self._commit_pending()

here is _commit_pending:

def _commit_pending(self):
    self.event_service.commit_bulk()
    self.inc_statistic(self.STATS_CNT_COMMIT_TIMEOUT, self.events_uncommitted)
    self.inc_statistic(self.STATS_CNT_COMMITS_TIMEOUT)
    self.events_uncommitted = 0
    self.last_commit = time.time()

and commit_bulk:

def commit_bulk(self):
    """ 
    Release and commit currently pending savepoint changes.
    """ 
    self.savepoint_release()
    self.commit()

Now in the function handling stopping of the storage:

def cbk_event_stop(self, daemon, args):
    """ 
    Stop the component.
    """ 
    daemon.logger.debug(
        "Component '{}': Stopping the component".format(
            self.cid
        )
    )
    # In case we are running in bulk commit mode.
    if self.commit_bulk:
        daemon.logger.info(
            "Component '{}': Committing all pending messages and switching to immediate commit mode.".format(
                self.cid
            )
        )
        # Commit all currently pending IDEA messages.
        self._commit_pending()
        # Switch to immediate commit mode for the rest of the messages in the queue.
        self._setup_insert_now(daemon)

one can see that _commit_pending is once again called resulting in calling savepoint_release twice.