Picture the following scenario: you typically run tablesnap under
supervisor. S3 fails (it has happened) or otherwise dies for some
reason. all your tablesnaps fail at more or less the same time, then
supervisor starts all your tablesnaps at the same time
The end result would be increased i/o util in all nodes of the cluster
at the same time, which is not good.
Now, I'm not saying is always a bad idea to compute md5 at start, if
you're doing it manually you can always do the initial load computing
md5 and then restart tablesnap without --md5-on-start (or configure it
in supervisor that way)
A couple more changed may have sneaked in, they are mostly for compatibility
with more packaging forms, tested and made sure they are working properly
Picture the following scenario: you typically run tablesnap under supervisor. S3 fails (it has happened) or otherwise dies for some reason. all your tablesnaps fail at more or less the same time, then supervisor starts all your tablesnaps at the same time
The end result would be increased i/o util in all nodes of the cluster at the same time, which is not good.
Now, I'm not saying is always a bad idea to compute md5 at start, if you're doing it manually you can always do the initial load computing md5 and then restart tablesnap without --md5-on-start (or configure it in supervisor that way)
A couple more changed may have sneaked in, they are mostly for compatibility with more packaging forms, tested and made sure they are working properly