OHDSI Home | Forums | Wiki | Github

Achilles performance in PostgreSql in Broadsea

Implementers
#1

I’ve installed Achilles v1.7.2 using the code below and Achilles is taking about 2 hours to run against a CDM with 2,500 patients running on PostgreSql in Docker using Broadsea on a Windows laptop with 64g memory (docker-compose.yml is attached).
docker-compose.yml.txt (6.2 KB)

remotes::install_github("OHDSI/Achilles@v1.7.2", INSTALL_opts = c("--no-multiarch"))

I’m seeing other posts reporting similar issues but not a resolution. Is this perfomance typical or is there something I can do to improve it?

https://forums.ohdsi.org/t/database-tuning-for-ohdsi/2568/2
https://forums.ohdsi.org/t/optimizing-omop-database-for-achilles/16655
https://forums.ohdsi.org/t/question-about-achilles-performance/16715

The script I’m running and the output is attached.
run-achilles.r.txt (1.1 KB)
output.txt (23.9 KB)

Thanks,
John

#2

Hi @greshje.gmail ,

For reference achilles runs in 50 seconds for me using broadsea with the demo cdm that’s includded (~2600 patients).

From your code it looks like you are not running on the demo cdm, is that correct? If it’s for example synthea data you created it might have a very beefy measurement table. If that one doesn’t have indexes that might cause slowness.

From the logs it looks like majority of the time is spent in few analyses that are taking 30+ min each. But we can’t see which one from the log.

Are you using broadsea-run-achilles to run achilles ?

Another way would be to run the script you attached outside docker, i.e. from a regular rstudio session. Then there should be a better log in “./output/log_achilles.txt” . At least that’s what happens when I do it (See attached).
log_achilles.txt (1.4 MB)

If you could get your hands on a log like that by running in rstudio we could see which analysis is taking all this time and proceed from there.

Hope this helps,
Egill

t