Closed Bug 1531378 Opened 6 years ago Closed 3 years ago

Windows GCP images

Categories

(Infrastructure & Operations :: RelOps: Windows OS, task)

task
Not set
normal

Tracking

(Not tracked)

RESOLVED WONTFIX

People

(Reporter: fubar, Assigned: grenade)

References

Details

Attachments

(2 files)

Once we have VPN connections up, we need to get our first Windows builder images running in GCP. We're focusing on builds first, so that we can test the artifacts in AWS. For the first go, we can build them manually from our AWS base and ripping out the AWS agents. Once they're working we need to automate them.

gecko-1-b-win2012-gamma now runs occ in gcp

still need to tweak the generic worker config to get it working for gcp (still have the configure-for-aws flag set)

Assignee: nobody → rthijssen
Status: NEW → ASSIGNED

some finger in the wind build timings (wip):

(In reply to Rob Thijssen [:grenade (EET)] from comment #19)

some finger in the wind build timings (wip):

Rob: have you tried the n1-standard-32 instance type? That's what we're using for Linux, and I'm curious whether the extra memory in the standard config has any impact vs highcpu for our build tasks.

n1-standard-32: 128GB memory, $0.32/hr
n1-highcpu-32: 28.80GB memory, $0.24/hr

In fact, this makes me want to retry our linux builds under that config too, given the price gap.

Flags: needinfo?(rthijssen)

(In reply to Chris Cooper [:coop] pronoun: he from comment #24)

In fact, this makes me want to retry our linux builds under that config too, given the price gap.

I ran this experiment this afternoon for Linux. Data is here: https://bugzilla.mozilla.org/show_bug.cgi?id=1525094#c13

In short, standard instances are faster, but may be overkill. However, since we want comparable performance to AWS for a lower price, we should absolutely find the GCE instance type config that gives us the required perf on Windows, and then go from there.

(In reply to Chris Cooper [:coop] pronoun: he from comment #24)

Rob: have you tried the n1-standard-32 instance type? That's what we're using for Linux, and I'm curious whether the extra memory in the standard config has any impact vs highcpu for our build tasks.

no. although i did try n1-highcpu-96 which has 86.4gb of ram (compared to n1-highcpu-32's 28.8) and saw negligible impact on build times.

testing n1-standard-32 now:

Flags: needinfo?(rthijssen)

logging here for a record of where i'm at with gcp windows 10 image imports...

windows 10 import:

gcloud beta compute images import windows-10-enterprise-1903-18362-30-en-us-x64 --source-file gs://windows-ami-builder/vhd/Windows_10_Enterprise_1903_18362_30_en-US_x64_MBR.vhd --os windows-10-byol --project image-builder-242310 --zone us-east1-b

currently erroring with:

[Daisy] Errors in one or more workflows:
  import-and-translate: step "import" run error: step "wait-for-signal" run error: WaitForInstancesSignal FailureMatch found for "inst-importer-import-and-translate-import-t0gtj": "ImportFailed: Failed to convert source to raw, error:     (0.00/100%)#015    (1.00/100%)#015    (2.00/100%)#015    (3.01/100%)#015    (4.01/100%)#015    (5.01/100%)#015    (6.01/100%)#015    (7.01/100%)#015    (8.02/100%)#015    (9.02/100%)#015    (10.02/100%)#015    (11.02/100%)#015    (12.02/100%)#015    (13.03/100%)#015    (14.03/100%)#015    (15.03/100%)#015    (16.03/100%)#015    (17.03/100%)#015    (18.04/100%)#015    (19.04/100%)#015    (20.04/100%)#015    (21.04/100%)#015    (22.04/100%)#015    (23.05/100%)#015    (24.05/100%)#015    (25.05/100%)#015    (26.05/100%)#015    (27.05/100%)#015    (28.06/100%)#015    (29.06/100%)#015    (30.06/100%)#015    (31.06/100%)#015    (32.06/100%)#015    (33.07/100%)#015    (34.07/100%)#015    (35.07/100%)#015    (36.07/100%)#015    (37.07/100%)#015    (38.08/100%)#015    (39.08/100%)#015    (40.08/100%)#015    (41.08/100%)#015    (42.08/100%)#015    (43.09/100%)#015    (44.09/100%)#015    (45.09/100%)#015    (46.09/100%)#015    (47.09/100%)#015    (48.10/100%)#015    (49.10/100%)#015    (50.10/100%)#015    (51.10/100%)#015    (52.10/100%)#015    (53.11/100%)#015    (54.11/100%)#015    (55.11/100%)#015    (56.11/100%)#015    (57.11/100%)#015    (58.12/100%)#015    (59.12/100%)#015    (60.12/100%)#015    (61.12/100%)#015    (62.12/100%)#015    (63.13/100%)#015    (64.13/100%)#015    (65.13/100%)#015    (66.13/100%)#015    (67.13/100%)#015    (68.14/100%)#015    (69.14/100%)#015    (70.14/100%)#015    (71.14/100%)#015    (72.14/100%)#015    (73.15/100%)#015    (74.15/100%)#015    (75.15/100%)#015    (76.15/100%)#015    (77.15/100%)#015    (78.16/100%)#015    (79.16/100%)#015    (80.16/100%)#015    (81.16/100%)#015    (82.16/100%)#015    (83.17/100%)#015    (84.17/100%)#015    (85.17/100%)#015    (86.17/100%)#015    (87.17/100%)#015    (88.18/100%)#015    (89.18/100%)#015    (90.18/100%)#015    (91.18/100%)#015    (92.18/100%)#015    (93.19/100%)#015    (94.19/100%)#015    (95.19/100%)#015    (96.19/100%)#015    (97.19/100%)#015    (98.20/100%)#015    (99.20/100%)#015qemu-img: error while reading sector 18214912: Input/output error'"
ERROR
ERROR: build step 0 "gcr.io/compute-image-tools/daisy:release" failed: exit status 1

backlog cleanup

Status: ASSIGNED → RESOLVED
Closed: 3 years ago
Resolution: --- → WONTFIX
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: