[1.3.1] [leo_gateway] Large Object Put Handler could halt infinitely

[windkit]

Description

When doing a large object put benchmark, leo_gateway would eventually stop responding. Every subsequent requests would get time out.

Observation

~/s3cmd-1.6.1$ parallel -P 16 "./s3cmd put /dev/shm/testfile_0 s3://test/{} &>> log/put{}.log" ::: $(seq 1 256)

Interpretation

1. High bumps come from s3cmd calculating md5sum of the ~1GB file
2. low bumps with 30-second interval comes from timeout

Root Cause

leo_pod:checkin is not be executed when the gen_server:call failed after leo_pod:checkout.
The worker is not returned and the error is not handled for restart.

Related PR

leo-project/leo_gateway#60

[yosukehara]

Fixed this issue by leo-project/leo_gateway/pull/60