in-house CI hitting bandwidth limit for gclient sync
Reported by
aourh...@gmail.com,
Apr 12 2018
|
||
Issue descriptionWhat is the bandwidth limit for 'gclient sync' and is there a registration process to raise the limit? Steps to reproduce: (1) Run parallel CI builds for multiple platforms & arch (2) gclient sync... (3) bandwidth limit reached Expected result: gclient sync succeeds This error occurs instead: [0:26:11] Cloning into 'E:\j\workspace\8b684a44\src\chrome\test\data\perf\_gclient_canvas_bench_jyjaqb'... [0:26:11] remote: Short term bandwidth rate limit exceeded for IP XX.XXX.XX.XXX [0:26:11] fatal: protocol error: bad pack header
,
Apr 12 2018
If no credentials (e.g. .gitcookies) are present on the machine(s), then it will have a quota limit enforced on a per-IP basis. If multiple machines are all performing similar fetches from behind a NAT, then they will all be in a shared quota pool for the IP of that NAT, thus exhausting their quota much faster. If credentials *are* presented by the machine(s), then quota will be enforced on a per-account basis, and that quota limit will be much higher than the per-IP "anonymous" quota. Given that the message sent from the remote is that the rate limit is exceeded for an IP, my assumption is that credentials are not present on these machines. So register an account, get its credentials onto all the machines, and use that to fetch from googlesource. That will likely increase your quota sufficiently to resolve this issue. At this time, we do not make a practice of granting quota increases to accounts which are not controlled by Google. |
||
►
Sign in to add a comment |
||
Comment 1 by a...@chromium.org
, Apr 12 2018Components: Infra
Owner: aga...@chromium.org