[content-service] Use gsutil to upload and download backups #10626
Conversation
aledbf
force-pushed
the
aledbf/gs
branch
2 times, most recently
from
aa2b0c6 to
57835db
Compare
aledbf
force-pushed
the
aledbf/gs
branch
16 times, most recently
from
137a5c5 to
52b0111
Compare
aledbf
force-pushed
the
aledbf/gs
branch
4 times, most recently
from
5908dc8 to
293c1d4
Compare
roboquat
added
release-note-none
and removed
do-not-merge/release-note-label-needed
labels
aledbf
changed the title
| args := fmt.Sprintf(`gsutil -q -m %v\ | ||
| -o "GSUtil:parallel_composite_upload_threshold=150M" \ | ||
| -o "GSUtil:parallel_process_count=3" \ | ||
| -o "GSUtil:parallel_thread_count=6" \ |
There was a problem hiding this comment.
Six threads seem to request lots of CPU resources, I captured the below picture from gen51 cluster.
Should we keep the parameter as now or should we limit more CPU resources for gsutil?
There was a problem hiding this comment.
Should we keep the parameter as now or should we limit more CPU resources for gsutil?
We have already limited the number of simultaneous backups to three. Also, the trade-off we are doing here is less RAM and better transfer speeds with more CPU utilization (short period of time).
That said if you have an idea to improve this, please post it here
There was a problem hiding this comment.
I tried to use https://github.com/EmbarkStudios/gsutil. That should consume fewer resources, but I could get it working consistently.
|
I think this change deserve a release note. Code LGTM. |
roboquat
added
deployed: workspace
Description
Switch to gcloud gsutil cli to upload and downloads for GCS interaction with backups. This improves the speed and also removes the multipart complexity from the codebase.
Release Notes