fix(jemalloc): Fix the stats of jemalloc #236
Conversation
ahsanbarkati
requested review from
jarifibrahim,
manishrjain and
martinmr
as code owners
| fmt.Printf("[%d] Current Memory: %05.2f G. Increase? %v\n", | ||
| counter, float64(curMem)/float64(1<<30), increase) | ||
| var js z.MemStats | ||
| z.ReadMemStats(&js) |
There was a problem hiding this comment.
Do you need this in memtest?
There was a problem hiding this comment.
I was thinking that this can be something which we can use to verify that the stats are as expected.
There was a problem hiding this comment.
Reviewable status: 0 of 2 files reviewed, 2 unresolved discussions (waiting on @ahsanbarkati, @jarifibrahim, @manishrjain, and @martinmr)
contrib/memtest/main.go, line 114 at r1 (raw file):
Previously, ahsanbarkati (Ahsan Barkati) wrote…
I was thinking that this can be something which we can use to verify that the stats are as expected.
The memstats can be added on. But, our version of how much memory should be used should still be printed.
z/calloc_jemalloc.go, line 164 at r1 (raw file):
sz := unsafe.Sizeof(&epoch) C.je_mallctl( (C.CString)("epoch"),
Link to documentation about this?
|
contrib/memtest/main.go, line 114 at r1 (raw file): Previously, manishrjain (Manish R Jain) wrote…
Yep, that is still printed. |
The jemalloc stats are read by using the
mallctl()call, this call doesn't return the latest stats unless we refresh the ctl cache. We can get the latest stats after calling an epoch mallctl. This PR fixes this issue.Note: Calling an epoch mallctl is expensive. An epoch mallctl takes up all the locks as taken by a malloc call. We should use this judiciously.
This change is