Database disconnection is not recovered well by deployment/benchmark

[gusmith]

This issue is opened from observations when running 100 times 1M*1M linkage on a deployed service on k8s.
At one point, the database fails and restarts (see issue #407 ).

From his point, the whole deployment melt-down: the workers restarts a number of times (I cannot see the source log with their failure), the flask pod restarts.
Flask seems to be OK when the database is not available (raising a lot of exception, but that sounds expected). But at the end and without clear notice:

[2019-08-22 18:54:22 +0000] [1] [INFO] Handling signal: term
[2019-08-22 18:54:22 +0000] [10] [INFO] Worker exiting (pid: 10)
[2019-08-22 18:54:22 +0000] [11] [INFO] Worker exiting (pid: 11)
[2019-08-22 18:54:22 +0000] [12] [INFO] Worker exiting (pid: 12)

[hardbyte]

Might be useful to look at automated retrying with an exponential backoff before failure.