The roads I take...
KaiRo's weBlog
| Displaying recent entries in English and tagged with "explosiveness". Back to all recent entries |
May 16th, 2016
Tools I Wrote for Crash (Stats) Analysis
Now that I'm off the job that dominated my life (and almost burned me out) for the last years, I finally have some time again to blog. And I'll start with stuff I actually did for that job, as I still am happy to help others to continue from where I left.
The more fun part of the stability management job was actually creating new analysis - and tools. And those tools are still helpful to people working on crash analysis or crash stats analysis now - so as my last task on the job, I wrote some documentation for the tools I had created.
One of the first things I created (and which was part of the original job description when I started) was a prototype for detecting crash "explosiveness", i.e. a detector for crashes that are rising significantly in volume. This turned out to be quite helpful for me and others to use, and the newest reports of it are listed in my Report Overview. I probably should talk about it in more detail at some point, but I did write up a plan on the wiki for the tool, and the (PHP) code is on hg.m.o (that was the language I knew best and gave me the fastest result for a prototype). I had plans to port/rewrite it in python, but didn't get to it. Calixte, who is looking after most of "my" tools now, is working on that though, and I have already promised to review his work as a volunteer so we can make sure we have this helpful capability in better code (and hopefully better UI in the end) for future use.
In general, I have created one-line docs for all the PHP scripts I had in the Mercurial repository, and put them into the run-reports script that is called by a daily cron job. Outside of the explosiveness script, most of those have been obsoleted by Socorro Super Search (yay for Adrian's work and for the ElasticSearch backend!) nowadays.
Also, the scripts that generate the summed-up data for Are We Stable Yet dashboard and graphs (also see an older blog post discussing the graphs) have been ported to python (thanks Peter for helping me to get started there) - and those are available in the Magdalena repository on GitHub. You'll see that this repository doesn't just have more modern code, using python instead of PHP and the public Socorro API instead of private PostgreSQL access, it also has a decent README documenting what it and every script in it does.
The most important tools for people analyzing crash stats are in the Datil repository on GitHub (and its deployment on crash-analysis), though. I used all those 4 dashboards/tools daily in the last months to determine what to report to Release Managers and other parties, find out what we need to file as bugs and/or push to get fixed. Datil, like Magdalena, has good docs right in the repository now, readable directly on GitHub.
So, what's there?
Well, the before-mentioned "Are We Stable Yet" dashboard and graphs, for sure (see the longtermgraph docs for what graphs you can get and a legend of what the lines mean).
There's also a tool/prototype for "what's important" weighed top crash lists that I called "Top Crash Score", see the score docs for what it does and examples on how to use that tool.
And finally, I created a search query comparison tools that did let me answer questions like "which crashes happen more with or without multi-process support (e10s) being active?" or "which crashes have vanished with the new beta and which have appeared (instead)?" - which was incredibly helpful to me at least. Read the searchcompare docs for more details and examples.
I probably won't spend a lot of time with those tools any more, neither in usage nor in development, but I'm still happy about people using them, giving me feedback, and I'm also happy to review and merge pull requests that feel like making sense to me!
The more fun part of the stability management job was actually creating new analysis - and tools. And those tools are still helpful to people working on crash analysis or crash stats analysis now - so as my last task on the job, I wrote some documentation for the tools I had created.
One of the first things I created (and which was part of the original job description when I started) was a prototype for detecting crash "explosiveness", i.e. a detector for crashes that are rising significantly in volume. This turned out to be quite helpful for me and others to use, and the newest reports of it are listed in my Report Overview. I probably should talk about it in more detail at some point, but I did write up a plan on the wiki for the tool, and the (PHP) code is on hg.m.o (that was the language I knew best and gave me the fastest result for a prototype). I had plans to port/rewrite it in python, but didn't get to it. Calixte, who is looking after most of "my" tools now, is working on that though, and I have already promised to review his work as a volunteer so we can make sure we have this helpful capability in better code (and hopefully better UI in the end) for future use.
In general, I have created one-line docs for all the PHP scripts I had in the Mercurial repository, and put them into the run-reports script that is called by a daily cron job. Outside of the explosiveness script, most of those have been obsoleted by Socorro Super Search (yay for Adrian's work and for the ElasticSearch backend!) nowadays.
Also, the scripts that generate the summed-up data for Are We Stable Yet dashboard and graphs (also see an older blog post discussing the graphs) have been ported to python (thanks Peter for helping me to get started there) - and those are available in the Magdalena repository on GitHub. You'll see that this repository doesn't just have more modern code, using python instead of PHP and the public Socorro API instead of private PostgreSQL access, it also has a decent README documenting what it and every script in it does.
The most important tools for people analyzing crash stats are in the Datil repository on GitHub (and its deployment on crash-analysis), though. I used all those 4 dashboards/tools daily in the last months to determine what to report to Release Managers and other parties, find out what we need to file as bugs and/or push to get fixed. Datil, like Magdalena, has good docs right in the repository now, readable directly on GitHub.
So, what's there?
Well, the before-mentioned "Are We Stable Yet" dashboard and graphs, for sure (see the longtermgraph docs for what graphs you can get and a legend of what the lines mean).
There's also a tool/prototype for "what's important" weighed top crash lists that I called "Top Crash Score", see the score docs for what it does and examples on how to use that tool.
And finally, I created a search query comparison tools that did let me answer questions like "which crashes happen more with or without multi-process support (e10s) being active?" or "which crashes have vanished with the new beta and which have appeared (instead)?" - which was incredibly helpful to me at least. Read the searchcompare docs for more details and examples.
I probably won't spend a lot of time with those tools any more, neither in usage nor in development, but I'm still happy about people using them, giving me feedback, and I'm also happy to review and merge pull requests that feel like making sense to me!
By KaiRo, at 22:33 | Tags: analysis, CrashKill, explosiveness, Mozilla, stability | no comments | TrackBack: 0