FWIW, I just looked and the longest job in the release steps so far is 8 minutes. Most are under 2 minutes. There might be jobs later that take longer that we haven't run yet. IMO, the issue isn't speed of the machine, it is just that we are sharing the machine with longer jobs (1 hour for TourDeFlexMigration). And again, the machine will be idle for stretches of time while the RM verifies artifacts after each step.
-Alex On 4/12/20, 11:32 AM, "Harbs" <[email protected]> wrote: Fair enough. I expect builds to be somewhere between 10 and 20 times faster on a powerful machine. Yeah. It’s probably going to be a bit of work changing the server, but probably worth it in the long run. I think I’ll try this when I do the next release unless Yishay wants to work with me on this for this release — but I’m not going to be able to help until after Passover (i.e. next week). Thanks, Harbs > On Apr 12, 2020, at 9:22 PM, Alex Harui <[email protected]> wrote: > > The Azure portal says: Standard F2s_v2 (2 vcpus, 4 GiB memory) > > I think I am reading changes to the build process in your suggestions. I do not really want to spend more of my time on this process. But if you want to do the work, that's fine with me. > > -Alex > > On 4/12/20, 10:57 AM, "Harbs" <[email protected] <mailto:[email protected]>> wrote: > > What kind of horsepower is included in the free Azure account? > > The server I mentioned builds (considerably) faster than my own local machine. The ci server seems to build many times slower. > > One thing we can do to minimize running server time would be to transfer the artifacts to storage instead of keeping them on the server. On AWS, I’d probably use S3. Not sure what the similar service on Azure is called. > >> On Apr 12, 2020, at 8:26 PM, Alex Harui <[email protected]> wrote: >> >> OK, that's pretty much how I understand Azure as well. The key thing is that "running" includes time where the CI server is not running any Jenkins jobs. The CI Server steps might take only a few hours of actual server time, but there is time where the RM is verifying artifacts locally so you'd be paying for that or the RM would have to keep shutting down and restarting. >> >> Seems like it would be cheaper/simpler to get the free MSDN account and leave it running. >> >> -Alex >> >> On 4/12/20, 10:15 AM, "Harbs" <[email protected]> wrote: >> >> My experience is with AWS. >> >> I assume Microsoft has similar offerings, but I don’t have experience with Azure. >> >> AWS has on-demand EC2 instances which you pay for only the actual time that they are running.[1] >> >> Instances can be started and stopped via command line (or via the web interface) as long as you have valid credentials to do so. >> >> For example: an m5.4xlarge instance has 16 cores and costs about $1.5 per hour. On a machine like that, a full build would probably take less than 10 minutes. It’s probably possible to do a full release with only a few hours of server time. >> >> Leaving a server like that running all the time would get expensive, but if it’s just spun up for releases, you’d get very fast builds at a reasonable price. >> >> I’d be happy to pay $10-$50 (and possibly more) per release to make the release process painless for the RM. >> >> [1]https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Faws.amazon.com%2Fec2%2Fpricing%2Fon-demand%2F&data=02%7C01%7Caharui%40adobe.com%7Cd99d1960963240ce500c08d7df0fe417%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637223131651425591&sdata=HPLHT0r9qTmJhR4f4j52wtslRJaCCnqf2lj8CM3x0LE%3D&reserved=0 <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Faws.amazon.com%2Fec2%2Fpricing%2Fon-demand%2F&data=02%7C01%7Caharui%40adobe.com%7Cd99d1960963240ce500c08d7df0fe417%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637223131651425591&sdata=HPLHT0r9qTmJhR4f4j52wtslRJaCCnqf2lj8CM3x0LE%3D&reserved=0><https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Faws.amazon.com%2Fec2%2Fpricing%2Fon-demand%2F&data=02%7C01%7Caharui%40adobe.com%7Cd99d1960963240ce500c08d7df0fe417%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637223131651425591&sdata=HPLHT0r9qTmJhR4f4j52wtslRJaCCnqf2lj8CM3x0LE%3D&reserved=0 <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Faws.amazon.com%2Fec2%2Fpricing%2Fon-demand%2F&data=02%7C01%7Caharui%40adobe.com%7Cd99d1960963240ce500c08d7df0fe417%7Cfa7b1b5a7b34438794aed2c178decee1%7C0%7C0%7C637223131651425591&sdata=HPLHT0r9qTmJhR4f4j52wtslRJaCCnqf2lj8CM3x0LE%3D&reserved=0>> >> >>> On Apr 12, 2020, at 7:45 PM, Alex Harui <[email protected]> wrote: >>> >>> I'm not very experienced with spinning up servers. The CI server we are using is effectively free, based on a generous donation from Microsoft of MSDN accounts to ASF committers. So I leave it up 24/7, and share the RDP access on private@. I think any other ASF committer could do the same. IIRC, if that server actually is stopped, I have to use my personal (unshared) MSDN credentials to start it again. AIUI, if I actually paid for the server, it would cost me to leave it running even if it didn't run jobs between releases. >>> >>> Is that what you are basically saying? I think it might be best if another committer got a CI server going via the MS donation and could leave it up 24/7. >>> >>> -Alex >>> >>> On 4/12/20, 9:28 AM, "Harbs" <[email protected]> wrote: >>> >>> I’m willing to do this. >>> >>> Considering that the release will be run infrequently, it should be doable to have a relatively powerful server that could be spun up on demand. This is something I have setup for my own releases. >>> >>> The only complication would be that each RM would need valid credentials to spin up the server. >>> >>> Harbs >>> >>>> On Apr 12, 2020, at 7:10 PM, Alex Harui <[email protected]> wrote: >>>> >>>> A better solution, IMO, is for someone else to offer up a CI server only for release jobs.
