Preliminary sandboxfs support and performance measurements

[Julio Merino]

Hello everyone,

After many months of work, I am happy to announce that support for sandboxfs-backed builds is available in experimental fashion starting with Bazel 0.12.0.

I know many of you have previously raised concerns about the slowness of sandboxing so this can help. To that extent, I'd like to request your input in quantifying how much this helps your use cases and whether the improvements are sufficient for you to consider using sandboxing.

Please head to the blog post below to find some background information, the performance results we have gotten so far, and brief instructions on how to use this:

    https://blog.bazel.build/2018/04/13/preliminary-sandboxfs-support.html

Let me know how it goes in this thread, in that blog post, or even in private email... and don't hesitate to ask for, or even offer, help ;-)

Have a good weekend.

--

Julio Merino / http://julio.meroh.net/

[Ian O'Connell]

Did this change/support make it to 0.12.0? 

bazel 0.12.0 doesn't seem to have an option `--experimental_use_sandboxfs`

maybe i missed something though?

--
You received this message because you are subscribed to the Google Groups "bazel-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bazel-discuss+unsubscribe@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bazel-discuss/CANk-BFpHvNj4f5seqctRfLpBOtmGm7CPEPeyegUUuqeph33n-Q%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

[Julio Merino]

On Thu, Apr 19, 2018 at 2:07 PM Ian O'Connell <ia...@stripe.com> wrote:

Did this change/support make it to 0.12.0? 

bazel 0.12.0 doesn't seem to have an option `--experimental_use_sandboxfs`

Oh wow. I have no idea how I read things and really believed that this actually shipped in 0.12. But a second look tells me you are right. (I may have missed the "baseline" note in the release commit, but can't tell now.)

I'll fix the blog post. And sorry, you'll have to use a head build to try this out or wait for the next release.

Please let me know how it goes!

--

Julio Merino / go/jmmv / Blaze

[rodr...@google.com]

Now that 0.13.0 is available this can be tested more easily. On a Linux workstation (Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz, 48vCPUs) I tried each config three times for our polyglot project and recorded the fastest time:

no sandbox: 405s
symlink sandbox: 456s (+13%)
sandboxfs: 442s (+9%)

Timed with: bazel clean --expunge && time bazel build //...

The variations in results within a given configuration were 15-30 seconds so the difference between the two sandboxes is within the error bars.