Bug 4392 - [Performance] sstate.bbclass: turn off some cache files' generation
Summary: [Performance] sstate.bbclass: turn off some cache files' generation
Status: VERIFIED WONTFIX
Alias: None
Product: OE-Core
Classification: Build System, Metadata & Runtime
Component: core (show other bugs)
Version: 1.5
Hardware: x86 Multiple
: Medium enhancement
Target Milestone: 1.5 M2
Assignee: Robert Yang
QA Contact:
URL:
Whiteboard:
Depends on:
Blocks:
 
Reported: 2013-04-23 06:36 UTC by Robert Yang
Modified: 2013-10-03 15:00 UTC (History)
6 users (show)

See Also:
OS type for building Yocto: ---
Type of Regression: ---
Verified:
Documentation change: No (bug/feature does not impact docs)


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Robert Yang 2013-04-23 06:36:53 UTC
We save the following sstate cache files currently:
packagedata.tgz
packagedata.tgz.siginfo
package.tgz.siginfo
package.tgz
populate-sysroot.tgz
populate-sysroot.tgz.siginfo
populate-lic.tgz.siginfo
populate-lic.tgz
deploy-rpm.tgz
deploy-rpm.tgz.siginfo

The end user only needs populate-sysroot.tgz and deploy-rpm.tgz in theory, maybe we can add a configure variable to let the user turn off the other files' generation (also turn off the .siginfo) to save the build time and disk space.

We may add some variables like: (turn on all of generation by default)

SSTATE_MODE[siginfo] = "0"
SSTATE_MODE[packagedata] = "0"
SSTATE_MODE[package] = "0"
...
Comment 1 Richard Purdie 2013-04-23 07:23:59 UTC
The end user does at least need packagedata and populate-lic, otherwise those have to get rebuilt from scratch. The only data that is not strictly needed is package and that is required if you change package formats for example.

Before we start making things configurable, I'd like to see performance numbers about the speed improvements this would give. I'm not sure its worth adding extra code paths just for one sstate target.
Comment 2 Robert Yang 2013-04-23 08:25:22 UTC
(In reply to comment #1)
> The end user does at least need packagedata and populate-lic, otherwise
> those have to get rebuilt from scratch. The only data that is not strictly
> needed is package and that is required if you change package formats for
> example.
> 
> Before we start making things configurable, I'd like to see performance
> numbers about the speed improvements this would give. I'm not sure its worth
> adding extra code paths just for one sstate target.

Hi RP, thanks for the review, it's not easy to disable the generation of these files, I will try it later to get the performance numbers.
Comment 3 Stoicescu Cornel 2013-10-03 15:00:10 UTC
Setting to VERIFIED as per above 2 comments. Please open a new bug if necessary.