value will result in an extra lookup on the target shard, but will result in no rows being found, as expected. However, if we want to allow for values to be reused, then the presence of this row will prevent heartbeat frequency. This means that we cannot guarantee the skew alignment feature will work as expected in certain conditions. This could happen mainly while streaming from replicas with high replication the ongoing state does not need to be interpreted on the client. To start a VStream Copy user is expected to provide an empty gtid along with a list of tables to copy (essentially a LastTablePK list with

value will result in an extra lookup on the target shard, but will result in no rows being found, as expected. However, if we want to allow for values to be reused, then the presence of this row will prevent the ongoing state does not need to be interpreted on the client. To start a VStream Copy user is expected to provide an empty gtid along with a list of tables to copy (essentially a LastTablePK list with request to the only tablet. Note: VREngine’s queries are actually tablet agnostic. The user is expected to restrict their queries to the dbname of the tablet. This is not a good user experience. We should

value will result in an extra lookup on the target shard, but will result in no rows being found, as expected. However, if we want to allow for values to be reused, then the presence of this row will prevent heartbeat frequency. This means that we cannot guarantee the skew alignment feature will work as expected in certain conditions. This could happen mainly while streaming from replicas with high replication the ongoing state does not need to be interpreted on the client. To start a VStream Copy user is expected to provide an empty gtid along with a list of tables to copy (essentially a LastTablePK list with

(roughly) a 1 minute interval between a running migration’s liveness reports. When a migration is expected to be running, and does not have a liveness report for 10 minutes, then it is considered stale. value will result in an extra lookup on the target shard, but will result in no rows being found, as expected. However, if we want to allow for values to be reused, then the presence of this row will prevent heartbeat frequency. This means that we cannot guarantee the skew alignment feature will work as expected in certain conditions. This could happen mainly while streaming from replicas with high replication

when replication lag is high and advise applications to hold the next operation. Applications are expected to break down their tasks into small sub-tasks. For example, instead of deleting 1,000,000 rows throttler does not implement this at this time. • 429 (Too Many Requests): Do not write. A normal, expected state indicating there is replication lag. This is the hint for applications or clients to withhold an alert mechanism if anything were to go wrong. Critical failures The following errors are not expected to happen. If they do, it means that 2PC transactions have failed to commit atomically: • InternalErrors

an alert mechanism if anything were to go wrong. Critical failures The following errors are not expected to happen. If they do, it means that 2PC transactions have failed to commit atomically: • InternalErrors from one shard to another. A workaround is to perform a delete followed by insert, which works as expected. Delete If the table owns lookup vindexes, then the rows to be deleted are first read and the associated includes: 1. MySQL error codes that we don’t explicitly handle. 2. MySQL response that wasn’t as expected. For example, we might expect a MySQL timestamp to be returned in a particular way, but it wasn’t

alert mechanism if anything were to go wrong. 67 Critical failures The following errors are not expected to happen. If they do, it means that 2PC transactions have failed to commit atomically: • InternalErrors from one shard to another. A workaround is to perform a delete followed by insert, which works as expected. Delete If the table owns lookup vindexes, then the rows to be deleted are first read and the associated systems up-to-date with the source. The VReplication feature itself is a fairly low level one that is expected to be used as a building block for the above use cases. However, it’s still possible to directly

alert mechanism if anything were to go wrong. 71 Critical failures The following errors are not expected to happen. If they do, it means that 2PC transactions have failed to commit atomically: • InternalErrors includes: 1. MySQL error codes that we don’t explicitly handle. 2. MySQL response that wasn’t as expected. For example, we might expect a MySQL timestamp to be returned in a particular way, but it wasn’t systems up-to-date with the source. The VReplication feature itself is a fairly low level one that is expected to be used as a building block for the above use cases. However, it’s still possible to directly

select a, b, c from t where x = :redacted1 and y = :redacted1 and z = :redacted2 and foo = 0x1337 Expected output: select a, b, c from t where x = :redacted1 and y = :redacted1 and z = :redacted2 and foo