Skip to main content

PyCon: Make Sure Your Programs Crash

See the website.

This talk was given by Moshe Zadka from VMware.

Think about how to crash and then recover from the crash.

If your application recovers quickly, stuff can crash and no one will see.

Even Python code occasionally crashes due to C bugs, untrapped exceptions, infinite loops, blocking calls, thread deadlocks, inconsistent resident state, etc. These things happen!

Recovery is important.

A system failure can usually be considered to be the result of two program errors. The second error is in the recovery routine.

When a program crashes, it leaves data that was written in an arbitrary program state.

Avoid storage: caches are better than master copies.

Databases are good at transactions and at recovering from crashes.

File rename is an atomic operation in modern OSs.

Think of efficient caches and reliable masters. Mark cache inconsistency.

He seems to be skeptical of the ACID nature of MySQL and PostgreSQL. I'm not sure why.

Don't write proper shutdown code. Always crash so that your crash code always gets tested. Your data should always be consistent.

Availability: if the data is consistent, just restart.

To get into the high 9s, recover very quickly. Limit impact, detect the crash quickly, and startup quickly.

Vertical splitting: different execution paths, different processes. Apache can have a child process die with no impact on availability.

Horizontal splitting: different code bases, different processes.

Watchdog: monitor -> flag -> remediate.

Watchdog principle: keep it simple, keep it safe.

A process can touch a file every 30 seconds. The watchdog sees whether the file has been touched.

The watchdog and the processor restarter should not be in the same process, because the watchdog should be simple. Remember: separation of concerns.

Mark problems. Check solutions. See if restarting worked.

Everything crashes: plan for it.

Linux has a watchdog daemon. Use that to watch your watchdog.


Anonymous said…
It's this kind of bullshit that makes me want to move on from python to something like Scala. Compare:

Q. How do we make software reliable?

Python A: make sure your code recovers quickly after crashing

Scala A: use Software Transactional Memory, to mark a block as a single transaction. Simply undo the current transaction if anything goes wrong, then continue from that point.

Q. How do we write programs that scale to high performance on multiple cores?

Python A: Well, we have the GIL which prevents proper multithreading, and actually makes threads SLOWER on multicore machines, but removing the GIL is hard, even though Jython did it just fine, so we're not going to bother. In short, we have lots of libraries for green threads etc., but none of them work because we won't fix the fundamental problems underlying it all.

Scala A: We solve the problem on two fronts. First, we the best of breed Actor model of concurrency, unifying it into the language core, with two main keywords allowing you to choose between heavyweight OS threads and lightweight "green" threads, allowing you to code in the same style no matter which system you need for a particular block of code. Secondly, we provide parallel collections, which allow you to, for instance, iterate over a list, running a block of code on it in parallel, just as you would with non-parallel code.

Python is being quickly left behind.
Anonymous, please don't use the word "bullshit" on my blog.

If you're interested in integrating STM into Python, see this (

I don't think that having code to recover from crashes quickly is at odds with other approaches to reliable software. For instance, Erlang has an actor-based concurrency model, and Erlang OTP is all about trees of actors that monitor each other and recover from failures quickly.
jj, I think you've got an extra a in Moshe's name.
Fixed! Thanks!