Gerrit Analytics applied to Android source code

0
Gerrit User Summit 2019 – GerritForge Inc. – Sunnyvale CA GerritForge.com 0
Gerrit Code Analytics
for the Android OpenSource Project
Luca Milanesio
Gerrit Code Review Maintainer
GerritForge

1
About GerritForge
Founded in
the UK
HQ in
London
Committed to
OpenSource
+ Sunnyvale
CA

2
Gerrit DevOps Analytics
§ There’s a lot value in your DevOps
pipeline
§ Information collected from Git, Jenkins,
Jira, you name it
§ Discover and publish meaningful KPI to
make intelligent decisions about
§ People
§ Projects
§ Infrastructure
§ Lower the Risk of a software release
leveraging insights on historical data

3
Continuous Delivery
Analytics Dimensions
People Reviews Projects Commits System Metrics

4
BigData to the rescue
§ Collect all review events
§ Collect all logs
§ Channel them to a central store
§ Crunch and Crunch continuously
§ Never delete
§ Process, inspect and learn

5
GDA Components
The main components of GDA are:
§ GDA Event Collector (Plugin)
This allows for data to be extracted, anonymized
and sent over to the next phase.
§ GDA ELT Engine
This is hosted in the cloud by GerritForge or on-
premises and functions as data mart and
processing for all development related data
§ GDA Dashboard(s)
These are provided by GF according to the
customer needs. Some dashboards are already
available (for people and projects). Others will
be built on purpose.

6
Android OpenSource Project use-case
Issue 10597: AOSP Gerrit stats page
Reported by zoran.jovanovic@sony.com on Wed, Mar 13, 2019,
9:02 AM PDT
AOSP repository and its Gerrit Code Review are a treasure trove of
data.
There are some very interesting and useful stats that could be
presented
to the users. The stats would help in giving recognition for the
contributors and reviewers alike, thus raising the motiviation of
contributors, it would provide an easy access to the long and rich
history of AOSP project etc.

7
The problems to be resolved
P1: How to mirror the AOSP repositories in a
systematic way?
P2: How to scale up the current Gerrit analytics
plugin + ETL?
P3: How to parse foreign Gerrit note-db data?

8
Problem 1: AOSP repositories replication
Gerrit has a replication plugin:
• Define replication remotes
• Define push ref-spec
How to pull instead?
Do you replicate the AOSP repos on your
Gerrit?
How do you automate the process?

9
Problem 1 solved: pull-replication plugin

10
Challenges with pull-replication plugin
1. Why not extending the replication plugin?
• replication.config already defines replication remotes
• Can the push logic be extended?
2. Why not developing a brand-new plugin?
• Copy & paste existing code, keeping the essential?
… or a mix of 1. and 2. ?

11
Design decisions
Refactoring of replication plugin
• Decouple configuration from remote destination
• Started in April, merged in October (6 months!)
• Introduced unit and integration tests (the firsts in 7 years !)
Reuse of the replication plugin logic in the pull-replication
plugin
• Use replication-plugin as a dependency
• Complete reuse of the configuration and logic associated

12
pull-replication demo
$GERRIT_SITE/etc/replication.config
[remote "aosp"]
url = https://android.googlesource.com/${name}
fetch = +refs/heads/*:refs/heads/*
fetch = +refs/tags/*:refs/tags/*
projects = platform/system/core

13
pull-replication demo
ssh localhost pull-replication start --all
==> logs/pull_replication_log <==
[2019-11-16 23:59:22,712] [49829e49] Replication from
https://android.googlesource.com/platform/system/core started...
[2019-11-16 23:59:22,753] [49829e49] Fetch references
[+refs/heads/*:refs/heads/*, +refs/tags/*:refs/tags/*] from
https://android.googlesource.com/platform/system/core
[2019-11-16 23:59:25,243] [49829e49] Replication from
https://android.googlesource.com/platform/system/core completed in 2530ms,
15012ms delay, 0 retries

14
Problem 2: AOSP repositories are BIG
Gerrit analytics plugin slowest
points:
• Processing of branches
• Binary files
• Commits diffs and stats
computation

15
Problem 2: solved
Gerrit analytics plugin performance
improvements:
• Pre-computation of branches
• Reuse of the Gerrit diff-cache for analytics
• Introduction of the analytics-
commits_statistics_cache

16
Analytics performance demo
Gerrit cold start (no cache):
$ time curl -v
'http://localhost:8080/projects/platform%2Fsystem%2Fcore/analyt
ics~contributors' | wc –l
1732
curl -v 0.02s user 0.03s system 0% cpu 1:30.60 total

17
Analytics added cache and performance
$ ssh localhost gerrit show-caches | grep analytics
analytics-commits_statistics_cache| 54011 | 1.6ms | 0% |
$ time curl -v
'http://localhost:8080/projects/platform%2Fsystem%2Fcore/analytics~contributors
' | wc –l
1732
curl -v 0.01s user 0.02s system 2% cpu 1.045 total
90x times faster with commits-stats cache

18
Analytics bot detection
People or BOTs can be recognized by patter of
commits:
• Detect file type by reg-ex
• Identify commits with BOT-like files only
$GERRIT_SITE/etc/analytics.config:
[contributors]
botlike-filename-regexp = OWNERS

19
Problem 3: foreign changes processing
Gerrit changes in NoteDb have a Server-ID baked-in
Example:
$ git show refs/changes/00/100000/meta
commit 0014ca6443ac0af338e2677b45e538782bb7a12e (origin/00/100000/meta,
refs/changes/00/100000/meta)
Author: beckysiegel <1030207@173816e5-2b9a-37c3-8a2e-48639d4f1153>
Date: Sat Mar 3 18:38:17 2018 +0000
Update patch set 1
Hashtag added: enhancement
Patch-set: 1
Hashtags: enhancement
Tag: autogenerated:gerrit:setHashtag

20
Solution: allow parsing of foreign NoteDb
Dec 2018: proposed - 9 months later, rejected and
abandoned

21
Solution revamped: Dave shows another way
Aug 2019: proposed - 1 month later, merged! (thanks, DavidO)

22
Analytics deployment on GerritHub
Gerrit master
(review-3)
World
Traffic (R/W)
Gerrit master
(review-4)
HAproxy HAproxy
Gerrit master
(review-1)
Gerrit master
(review-2)
HAproxy HAproxy
Analytics Traffic (R/W)
Multi-site plugin
Multi-site plugin
Multi-site plugin
Multi-site plugin

23
Analytics next steps
1. Enable pull-replication on review-
{3,4}.gerrithub.io
2. Process AOSP repos contributors analytics
once-a-day
3. Publish commits stats to
analytics.gerrithub.io
4. Enrich with hashtags change data

24
Poll: have you implemented multi-master/HA?
Image from: http://cypp.rutgers.edu/ru-voting/political-information/public-opinion-polls/

25
Wants to know more?
GerritForge.com/contact

Gerrit Analytics applied to Android source code

More Related Content

What's hot

Similar to Gerrit Analytics applied to Android source code

More from Luca Milanesio

Recently uploaded

In this document

Gerrit Analytics applied to Android source code