storage: decrease default kv.bulk_io_write.max_rate#36884
storage: decrease default kv.bulk_io_write.max_rate#36884craig[bot] merged 1 commit intocockroachdb:masterfrom
Conversation
|
This change is |
7c0e18e to
f85705e
Compare
tbg
left a comment
There was a problem hiding this comment.
I was tempted to ask you to validate the byte setting, but it would be unclear what the validation is (other than saying it can't be more than 2TB or something like that). What was the underflow? A float64 can represent 1/MaxInt64 but I assume it did more math with it than that. Is that rate limiter generally wonky (as in, numerically unstable or the like)?
|
I just found #36806 (comment). I like to link to discussions when submitting a PR, consider doing that here too. |
Previously `kv.bulk_io_write.max_rate` had a default value of `MaxInt64`, which caused rounding problems in the rate limiter due to very small time intervals. This lowers the default to 1 TB/s. Release note (bug fix): The default value of `kv.bulk_io_write.max_rate` is now 1 TB/s, to help prevent incorrect rate limiting behavior due to rounding.
f85705e to
8f53788
Compare
|
"Underflow" isn't the right word. I thought it might have been part of the problem originally, but it's actually just a more straightforward case of an integer cast rounding a small float to 0. (I updated the commit message.) Regarding validation, I expect that nobody will try to raise the setting past 1 TB - the goal here was just to make a minimal change so the default setting won't get the rate limiter into a bad state, but I'm open to adding it. |
|
We should hold off on adding validation if we want to back port this. |
|
bors r+ |
36884: storage: decrease default kv.bulk_io_write.max_rate r=lucy-zhang a=lucy-zhang Previously `kv.bulk_io_write.max_rate` had a default value of `MaxInt64`, which caused rounding problems in the rate limiter due to very small time intervals. This lowers the default to 1 TB/s. Fixes #36806. (See #36806 (comment)) Release note (bug fix): The default value of `kv.bulk_io_write.max_rate` is now 1 TB/s, to help prevent incorrect rate limiting behavior due to rounding. Co-authored-by: Lucy Zhang <lucy-zhang@users.noreply.github.com>
Build succeeded |
Previously
kv.bulk_io_write.max_ratehad a default value ofMaxInt64, whichcaused rounding problems in the rate limiter due to very small time intervals.
This lowers the default to 1 TB/s.
Fixes #36806.
(See #36806 (comment))
Release note (bug fix): The default value of
kv.bulk_io_write.max_rateis now1 TB/s, to help prevent incorrect rate limiting behavior due to rounding.