Jira Link: DB-1640
If a tserver goes away for more than 5 minutes and then comes back, it will have been kicked out of its Raft groups by then. The master will then assign some tablets to it and do remote bootstrap. However, it is likely that the data the tablet server had is not too different from the new data for the tablets that it had, especially for large data volumes. We could try to reuse the data in old files and just extract the necessary delta and install it as a new set of RocksDB files. This might be complicated by garbage-collection of delete markers for some rows that the failed server had that has already happened in live servers while this server was down.
Jira Link: DB-1640 If a tserver goes away for more than 5 minutes and then comes back, it will have been kicked out of its Raft groups by then. The master will then assign some tablets to it and do remote bootstrap. However, it is likely that the data the tablet server had is not too different from the new data for the tablets that it had, especially for large data volumes. We could try to reuse the data in old files and just extract the necessary delta and install it as a new set of RocksDB files. This might be complicated by garbage-collection of delete markers for some rows that the failed server had that has already happened in live servers while this server was down.