Unidentified thread created in FlushStateToDisk() #19199

While testing #19197, a strange, at least for me, behavior has been observed during reindexing with:

$ ./src/bitcoind -testnet -reindex

When the dbcache reaches its maximum and the following lines appear in the log:

2020-06-07T13:24:37Z [loadblk] Cache size (761427872) exceeds total space (761373440)
2020-06-07T13:24:42Z [loadblk] FlushStateToDisk: write coins cache to disk (5389943 coins, 761427kB) started
2020-06-07T13:24:50Z [loadblk] FlushStateToDisk: write coins cache to disk (5389943 coins, 761427kB) completed (7.60s)

a new thread has been created by the client.

The strange thing is that this thread is named "loadblk". I understand that it couldn't be the ThreadImport thread as it is guarded by CImportingNow instance.

In my tests this unidentified thread is mostly named "loadblk", but once I've noted "msghand" name.

It seems this thread is unjoined until the client shutdown.

Some screenshots:

System: x86_64, Linux Mint 19.3 (Ubuntu 18.04 code base)

Does leveldb start threads?

Yeah, I assume it's one of the LevelDB threads. I vaguely remember a PR recently to rename it.

Does leveldb start threads?

Yes, it does. From the leveldb docs:

When the size of level L exceeds its limit, we compact it in a background thread.

I vaguely remember a PR recently to rename it.

#18790. But the "Name leveldb thread with its own name" (d8e3cf858c388fc15b5d5a506e2fa551d1fc0e07) commit was dropped.

I verified. It is a leveldb background thread.