Retrying instead of Reconstructing Snapshot in case of "Error reading snapshot"

Imported from Google Groups
#1

Hi!

We, let’s say once a day, have the phenomenon the we see the following error in our logfiles:

Error reading snapshot. Reconstructing aggregate from entire event stream. Caused by: javax.persistence.PersistenceException org.hibernate.exception.GenericJDBCException: could not extract ResultSet

Since our aggregate is constructed out of hundreds and thousands of events, simply reconstructing takes quite a long
time. I have the feeling that the error has something to do with database rollbacks and
therefore simply retrying should bring the snapshot up again.

So my question: Is there a way of retrying instead of completely rebuilding a snapshot?

#2

Hi Dirk,

That’s bothersome…and I think a correct assumption.

A simply retry on the reading of the snapshot might proof beneficial in your scenario.

I checked whether the current set up allows to retry reading a snapshot, but that’s sadly not the case as far as I can see.

What you could do, is introduce your own implementation of the [JPA/JDBC/Mongo]EventStorageEngine and override the readSnapshotData(String) function to incorporate a retry mechanism.

Alternatively, you could of course add an issue/feature request here, and let us do the work.

Or, introduce a PR introducing that functionality!

In short, it’s not available atm, but shouldn’t be to difficult to add into the framework or your personal project.

Hope this helps!

Cheers,

Steven

#3

Hi Steven,

thanks for your reply. I will see what I can do!

#4

Hi Steven,

I’m wondering whether it would be a better Idea to fail fast when we are unable to read the snapshot.
E.g. like this in the AbstractEventStore Class:

public DomainEventStream readEvents(String aggregateIdentifier) {
    Optional<DomainEventMessage<?>> optionalSnapshot;
    try {
        optionalSnapshot = storageEngine.readSnapshot(aggregateIdentifier);
    } catch (Exception | LinkageError e) {
        String message = String.format("Error reading snapshot. I stop to read the events for aggregateIdentifier %s. Caused by: %s %s",
                aggregateIdentifier, e.getClass().getName(), e.getMessage());
        throw new JdbcException(message, e);
    }

If we have an exception due to a db rollback, a retry wouldn’t help since the transaction has already been marked as rollback only.
What do you think?

#5

As discussed in the issue (https://github.com/AxonFramework/AxonFramework/pull/607), we have added a commit that extracts this behavior to a protected method, allowing you to override the behavior if you wish to.
See commit: https://github.com/AxonFramework/AxonFramework/commit/ecf0ebfefafe2c1fa910d3acff1541ce8f258e01

Feel free to provide feedback or comments if this doesn’t fully resolve the issue.

Cheers,

Allard