rdhabalia opened a new pull request #13073:
URL: https://github.com/apache/pulsar/pull/13073
### Motivation
we need the support of ledger roll-over for inactive topics to prevent
data-loss. It happens when Inactive topics that do not have publish for a long
time and the current ledger segment ensemble bookies are replaced by replicator
when ensemble bookies went into maintenance mode. In this scenario, there is a
chance of data loss and rolling over such ledgers prevents data loss.
For example:
1. In the beginning topic is writing to ledger `1000` with bookie `b1` and
`b2`
```
get /ledgers/00/0000/L1000
BookieMetadataFormatVersion 2
quorumSize: 2
ensembleSize: 2
length: 0
lastEntryId: -1
state: OPEN
segment {
ensembleMember: "b1:3181" -> prod5-bookie7.messaging.gq1.yahoo.com
ensembleMember: "b2:3181" -> prod5-bookie3.messaging.gq1.yahoo.com
firstEntryId: 0
}
```
2. Then both bookie b1 and b2 went down and replicator replaced them with
bookie b3 and b4
```
get /ledgers/00/0000/L1000
quorumSize: 2
ensembleSize: 2
length: 104
lastEntryId: 1
state: CLOSED
segment {
ensembleMember: "b3:3181"
ensembleMember: "b4:3181"
firstEntryId: 0
}
segment {
ensembleMember: "b5:3181"
ensembleMember: "b6:3181"
firstEntryId: 1
}
```
3. after that bookie b1 and b2 came back again and topic received a publish.
broker was able to reconnect to bookie b1, and b2 and continued to write to
bookie b1 and b2
```
Broker-DEBUG log:
22:45:24.000 [bookkeeper-ml-workers-OrderedExecutor-6-0] DEBUG
org.apache.bookkeeper.mledger.impl.ManagedLedgerImpl -
[prop/us-west1/test/persistent/t1] Write into current ledger lh=291607
entries=14
# broker successfully connected to b1 and b2 and wrote new entry to b1 and b2
22:36:38.949 [bookie-io-1-6] INFO o.a.b.proto.BookieRequestHandler -
Channel connected [id: 0x43c62e1c, L:/10.10.10.10:3181 - R:/b1:60908]
22:36:38.950 [bookie-io-1-5] INFO o.a.b.proto.BookieRequestHandler -
Channel connected [id: 0x43c62e1c, L:/10.10.10.10:3181 - R:/b2:60908]
```
4. So, according to zk ledger `1000` has one entry but the broker is kept
writing to b1 and b2 bookies which can cause the data loss.
[Bookie PR#2616](https://github.com/apache/bookkeeper/pull/2616) can be
helpful in this scenario but rolling over inactive ledger helps to user who are
doing bookie replacement/maintenance more frequent to avoid kind of data loss.
### Modification
- added config `managedLedgerInActiveLedgerRolloverTimeMinutes` to enable
rollover of inactive ledgers (by default it will be 0)
### Result
this feature helps in case of inactive topics which write to replaced
bookies during any maintenance and helps to avoid data-loss in that situation.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
