Some update from my side.
NOTE: I’m not recommending this fix in any way. Still trying to understand all the implications of it. Probably it’s safe, but still don’t know for sure.
After spending few hours trying to understand what and why is happening, I decide to moved on with manually fix those “transactions” entries. However, in replicaset, config.transactions cannot be modified, so what I did was:
- Remove on of the replicas from the set and start it in standalone mode.
- Delete the 2 entries.
- Restart again as part of the replicaset (
--replSet rs0flag) and add it back. Wait to replicate changes. - Do the same with the second replica.
- Stepdown the primary and do the same.
After that, cluster was able to upgrade to v4.2 FCV and looks like working correctly and without any downtime.
Still, I’m open if there is other solutions for this problem and any feedback regarding above.