540ba9dde0
I added document explaining ninjalog collection. This document explains what type of data is collected and how the data will be used. Bug: 900161, b/120645342 Change-Id: If47517aaa7a5510e4a634b5ccc0e4127d014dcb9 Reviewed-on: https://chromium-review.googlesource.com/c/1381912 Commit-Queue: Takuto Ikuta <tikuta@chromium.org> Reviewed-by: Dirk Pranke <dpranke@chromium.org>
1.9 KiB
Ninja build log collection
[TOC]
Overview
When chromium developers use autoninja for their build,
e.g.
$ autoninja -C out/Release chrome
autoninja uploads ninja's build log for Google employees but we don't collect logs from external contributors.
We use this page to decide whether an autoninja user is a Googler or not. Only people with access to Google's internal network can see 'Success'.
Before uploading logs, autoninja shows a message 10 times to warn users that we will collect build logs.
autoninja users can also opt in or out by using the following commands:
ninjalog_uploader_wrapper.py opt-inninjalog_uploader_wrapper.py opt-out
What type of data are collected?
The collected build log contains
- output file of build tasks (e.g. chrome, obj/url/url/url_util.o)
- hash of build command
- start and end time of build tasks
See manual of ninja for more details.
In addition to ninja's build log, we send the following data for further analysis:
- OS (e.g. Win, Mac or Linux)
- number of cpu cores of building machine
- build targets (e.g. chrome, browser_tests)
- parallelism passed by -j flag
- following build configs
- host_os, host_cpu
- target_os, target_cpu
- symbol_level
- use_goma
- is_debug
- is_component_build
We don't collect personally identifiable information (e.g. username, ip address).
Why ninja log is collected? / How the collected logs are used?
We (goma team) collect build logs to find slow build tasks that harm developer's productivity. Based on collected stats, we find the place/build tasks where we need to focus on. Also we use collected stats to track chrome build performance on developer's machine. We'll use this stats to measure how much we can/can't improve build performance on developer's machine.