Github user suyogmapara commented on the issue:
https://github.com/apache/zookeeper/pull/567
Hi @maoling
Regarding 2, I agree that this would be the outcome. But sorry it is not
clear to me why do you think that the result is surprising. You are setting
log-size-limit slightly larger than the pre-alloc size and so you would see
that actual log size is twice the preAllocSize.
Regarding 1, it seems to me that we configured zookeeper to perform poorly
and so it would do poorly. It is similar to setting snapCount = 1.As I said
above, I can explicitly prevent this case if you guys think such a validation
is useful.
As a side note, example of a good config would be preAllocSize = 64 MB,
txnLogSizeLimit = 512 MB.
Regarding 3, 1 MB is a quite large and typical configuration of snapCount
would mean we will have txnLogs of several GBs.
Regarding 4, txnLogSizeLimitInKb controls maximum size of the transaction
log. I believe there is a separate parameter that controls batch commit size on
observers.
---
