roachprod: Minor readme updates #36101
Conversation
Use new convenience commands and demonstrate running a workload Release note: None
|
This change is |
There was a problem hiding this comment.
LGTM
Reviewable status:
complete! 0 of 0 LGTMs obtained (waiting on @andreimatei and @bdarnell)
pkg/cmd/roachprod/README.md, line 38 at r1 (raw file):
# load generator for some tests. Note that the cluster name must always begin # with your username. export CLUSTER="${USER}-test"
say something about sourcing the bash completion script in ~/.roachprod (although the thing doesn't complete cluster names for me for some reason...)
pkg/cmd/roachprod/README.md, line 50 at r1 (raw file):
# ...or using roachprod directly (e.g., for your locally-built binary). build/builder.sh mkrelease roachprod put ${CLUSTER} cockroach-linux-2.6.32-gnu-amd64 cockroach
isn't the binary in bin.linux-amd64/cockroach-linux-... or so on Macs?
There was a problem hiding this comment.
Reviewable status:
There was a problem hiding this comment.
bors r=andreimatei
Reviewable status:
pkg/cmd/roachprod/README.md, line 38 at r1 (raw file):
Previously, andreimatei (Andrei Matei) wrote…
say something about sourcing the bash completion script in ~/.roachprod (although the thing doesn't complete cluster names for me for some reason...)
I'll leave that for someone who knows something about it (and who can get it to work as expected); I run stuff from emacs so I don't use custom bash completion anyway.
pkg/cmd/roachprod/README.md, line 50 at r1 (raw file):
Previously, andreimatei (Andrei Matei) wrote…
isn't the binary in
bin.linux-amd64/cockroach-linux-...or so on Macs?
Not with builder.sh mkrelease
|
bors r+ Looks like bors missed this before. |
|
bors ping |
|
pong |
|
bors r+ |
36101: roachprod: Minor readme updates r=bobvawter a=bdarnell Use new convenience commands and demonstrate running a workload Release note: None 37282: exec: add templating for RANK and ROW_NUMBER window functions r=yuzefovich a=yuzefovich Templates out rankOp into two operators (for dense and non-dense case) and templates out support for PARTITION BY clause for both rankOp and rowNumberOp. The code is undertested and is lacking benchmarks, but that will be addressed soon. 37412: server/heapprofiler: don't consider "goIdle" memory r=andreimatei a=andreimatei Before this patch we were taking a heap profile once RSS got above 85% of system memory and then we wouldn't take another profile until we go below it and then again above it. The problem with the RSS is that it includes "goIdle" memory (memory allocated from the OS that's not actually in use) ; I've seen it be up to several gigs. With this patch, we now take a profile every time a new high-water mark is reached. The recorded high-water mark is reset once an hour, this way ensuring that one big measurement doesn't stop us forever from taking another profile. It's simpler and it has a chance of catching some further heap increases. The rationale behind the patch is that we want profiles when the heap is large more than when the RSS is large. I'm looking at a case where we took a heap profile when the heap was 4.5 gigs with 2 gigs idle and then never took one again because of how the heuristic works. And then we OOMed when the heap was larger and the idle space was lower, but the RSS was about the same. With this patch, we would have taken a profile at a more interesting time. Release note: None Co-authored-by: Ben Darnell <ben@bendarnell.com> Co-authored-by: Yahor Yuzefovich <yahor@cockroachlabs.com> Co-authored-by: Andrei Matei <andrei@cockroachlabs.com>
Build succeeded |
Use new convenience commands and demonstrate running a workload
Release note: None