Closed kvantricht closed 4 weeks ago
Currently updating the executor-memory
and executor-memoryOverhead
greatly helped.
job_options={
"driver-memory": "4g",
"executor-memory": "3g",
"executor-memoryOverhead": "5g",
"udf-dependency-archives": [f"{ONNX_DEPS_URL}#onnx_deps"],
}
Further investigation revealed this issue: https://github.com/Open-EO/openeo-gfmap/issues/142
I also added a new config to better constrain python memory use: https://github.com/eu-cdse/openeo-cdse-infra/issues/44 And then made decommission less aggressive: https://github.com/eu-cdse/openeo-cdse-infra/issues/197
With these things combined, I could run a job with total executor memory of 4GB, for a 20x20km tile.
Currently any job beyond 5km X 5km fails with a confusing memory issue. This needs to be investigated and fixed before we can send this system for review to ESA.