Application Performance Monitoring in Tryton
0 likes411 views
The document discusses application performance monitoring (APM) for Tryton, highlighting the challenges in identifying and analyzing bugs and crashes in production. It introduces a patch called apm.py for profiling and logging RPC calls, including signal usage for performance tracking and error logging in JSON format. Additionally, it addresses data gathering methods to collect and store information from servers for further analysis.
Application Performance Monitoring in Tryton
- 2. 1/Why?
- 3. ● Usually crashes are pretty easy to fix ● Other kind of bugs 1.1/Bugs/crashes?
- 4. ● It is hard to analyze specially on production because it's always too late ● Lots of elements: ● Server hardware ● Virtualization & OS ● PostgreSQL ● Python ● Trytond ● Trytond Modules ● Network ● Client 1.2/Performance
- 5. ● Ideally: should take into account as much layers as possible ● But, let's concentrate on Tryton 1.3/Specialized instrumentation
- 7. ● A patch/hack on trytond/protocols ● Signals: ● SIGUSR1: Print currently running RPC calls in the log ● User ID ● Object & RPC method ● Postgres PID – APM! ● Current total execution time of the request ● SIGUSR2: Print backtrace in the log – great for finding performance issues 2.1/The patch: Signals
- 8. ● Profiling using: ● cProfile – A lof of overhead – should not be used in production (usually) ● vmprof (from pypy project) – 2~3% overhead ● When? ● Criteria in trytond.conf (should it be dynamic??) ● Criteria can include: user, rpc request ● Storage of the trace based on criteria (same as above + total time of request) – stores a Transaction ID 2.2/The patch: Profiling
- 9. ● Logging in JSON format: ● Transaction ID ● Time spent on transaction ● RPC request ● User ● Only log based on criteria in trytond.conf 2.3/The patch: Logging
- 10. ● Similar to Sentry ● Give the user a “Transaction ID” for tracking the error ● Maybe we should convert the Transaction ID into a docker- like string: adjective_name such as angry_einstein 2.4/The patch: Friendly user message for crashes
- 12. ● Python ● Send and store in JSON format ● Pick information from servers to RabbitMQ server ● Store that information and query from the command line 3.1/Recull
- 13. http://www.NaN-tic.com Albert Cervera i Areny albert@nan-tic.com @albertnan linkedin.com/in/albertca