Disabled compaction during reconnect leads to a large number of files/descriptors

[OlegMazurov]

Problem

When new data is received into a VirtualMap during reconnect it is regularly flushed but never compacted if the new status after reconnect is BEHIND. The number of files in the data base and open descriptors in the java process may grow very high:

# ls -l /proc/`pgrep -f java`/fd | wc -l 
76142

Solution

Possible solutions:

1. Revisit the decision to disable compaction in learner's database
2. During reconnect, we are in continuous flushing mode (even if working with multiple VirtualMaps concurrently). We could continue flushing a VirtualMap into the same set of files and close them once reconnect is finished.

Alternatives

No response

[OlegMazurov]

See also #9588