Avi Kivity

It depends on the branching factor (ratio of row size to key size) and workload. High branching factor and/or wide key read distribution: row cache is less effective, index cache...

Backported the flag to 4.6, 5.0, 5.1.

> os.join.path for figuring out non root nvmes > > fixes #8778 It doesn't mention why it uses glob instead of os.join.path. Also, interesting choice of where to end the...

@gleb-cloudius please review (from the point of view of someone that has to use the new metrics)

Please update the cover letter with information about why we want it, and what kind of tasks it will manage.

``` 600000000000-6001dc200000 rw-p 00000000 00:00 0 6001dc200000-601000000000 ---p 00000000 00:00 0 601000000000-6011dc200000 rw-p 00000000 00:00 0 6011dc200000-602000000000 ---p 00000000 00:00 0 602000000000-6021dc200000 rw-p 00000000 00:00 0 6021dc200000-603000000000 ---p 00000000 00:00...

Scylla tries to allocate 52 GiB: ``` >>> 0x1dc200000 * 7 / (1024**3) 52.076171875 ```

Please take a /proc/pid/maps snapshot when it grows by at least 1GB so we can see where it went.

Main heap usage is unchanged. ```console $ awk '/^[^6].*rw-p/ { split($1, a, "-"); s += strtonum("0x" a[2]) - strtonum("0x" a[1]) } END { print(s) }' scylla_maps_52_1_to_53_1_gb_increase.txt 33415168 ``` excess heap...

Please capture /proc/.../smaps instead, it has additional information. From the capture above, the heap looks fine, but maybe the memory leaked somewhere else.