Keep the Gradle user home of the Android lint tasks in a worker cache
Categories
(Firefox Build System :: Task Configuration, enhancement)
Tracking
(firefox161 fixed)
| Tracking | Status | |
|---|---|---|
| firefox161 | --- | fixed |
People
(Reporter: ahochheiden, Assigned: ahochheiden)
References
(Blocks 1 open bug)
Details
Attachments
(1 external file)
The Android lint tasks keep GRADLE_USER_HOME inside the checkout (mobile/android/gradle/dotgradle-offline), which run-task cleans before every task. Every task starts Gradle with no compiled build scripts, transforms or generated jars. In this autoland android-lints run the first Gradle run spends 39s between "Calculating task graph" and the first "Configure project":
https://firefox-ci-tc.services.mozilla.com/api/queue/v1/task/JdPa1UkURmGzc3Ou0PE-xA/artifacts/public/logs/live_backing.log
Task: https://treeherder.mozilla.org/jobs?repo=autoland&revision=7357050d4e9eaf9cf4e665ccea12f735a67f2eea&selectedTaskRun=JdPa1UkURmGzc3Ou0PE-xA.0
Moving it to a persistent worker cache lets later tasks on the same worker reuse that state. Measured on try, android-lints on workers with a warm checkout (medians):
- Gradle user home cold: 281s (8 runs).
- Gradle user home warm: 208s (7 runs), about 73s faster.
The export step drops from about 122s to 66s, and the first configuration from about 39s to 9s. The lint itself takes the same time.
| Assignee | ||
Comment 1•20 days ago
|
||
The Gradle user home lived in the checkout, which run-task cleans, so every
task started with no compiled build scripts, transforms or generated jars.
On a worker that keeps the cache, android-lints is about 73s faster (281s
to 208s on try).
Updated•20 days ago
|
Comment 3•18 days ago
|
||
| bugherder | ||
Description
•