intermittent issues with ClamAV clamd INSTREAM on

I've got an AWS Lambda function running NodeJS code to stream files from S3 to ClamAV running on an EC2 instance.

Generally (about 75% of the time) the system works, but often (especially when multiple files are being scanned from different Lambda containers) clamd threads gets stuck on INSTREAM.

Once a thread has been in INSTREAM for 25-30 seconds it does not seem to be able to recover. When it has been QUEUEDSINCE 350 seconds it is killed off. I can't figure out how either of these numbers relate to any value in my config.

I'm struggling to find any sign of an error in the logs - the number of INSTREAM requests matches the number of complete scans:

$ sudo grep -c "got command INSTREAM" /var/log/clamav/clamav.log
129
$ sudo grep -c "Chunks complete" /var/log/clamav/clamav.log
129
$ sudo grep -c "Scanthread: connection shut down" /var/log/clamav/clamav.log
129

...okay, now that I look a little more deeply into the logs it just takes a lot longer for some to be scanned. When I do a batch of 16 files, with Lambda concurrency restricted to 7 the first 7 files are scanned within a few seconds. The next file begins scanning soon after, gets to "Chunks complete" within a second, but takes 23 seconds before "Scanthread: connection shutdown". From here on it just gets worse - 1:24, 1:45... and then the 3rd batch of 7 files take over 3 minutes to scan.

If I give the system a few minutes to settle down, all the threads to die off, the same files that took over 3 minutes now take about 5-7 seconds.

If I run the same test on a faster machine the performance improves, but the issue is still there:

When threads get stuck at INSTREAM I can see that the files are still there:

$ ls -al /tmp
drwx------  2 clamav clamav     4096 Aug 29 16:52 clamav-493bdf893ce4d8d7763c00fee22d9d69.tmp
-rwx------  1 clamav clamav 25683921 Aug 29 16:52 clamav-5cdefd83d5531a03c7cf22fda37d133f.tmp