Open chrs-myrs opened 5 years ago
Hi @chrs-myrs - this can be an issue if you have a large number of source files that you're trying to convert. As a workaround, can you try setting spark.driver.maxResultSize
on the Glue job?
In the "Security configuration, script libraries, and job parameters (optional)". ===> "Job Parameters" section, add the following key --conf
and value spark.driver.maxResultSize=2g
.
Long term, we may need to find a way to better filter the initial set of inbound files to a smaller set, possibly as part of #12.
While adding spark.driver.maxResultSize=2g
or higher, it's also good to increase driver memory so that the allocated memory from Yarn isn't exceeded and results in a failed job.
The setting is spark.driver.memory
.
Adding two spark configs is done like this:
Key: --conf
Value: spark.driver.maxResultSize=2g --conf spark.driver.memory=8g
Setting the maxResultsSize gave us enough to get this to run properly
I'm experiencing this error, but only in subsequent job executions, the first time I run the job even with 100,000s of files in the processing folder (CloudFront logs) it will work with no memory issues. However on subsequent runs it keeps failing. Anyone got any idea? I've been trying move files around and process in batches, but it's a pain. Should this library be able to handle huge file numbers without issues? Or should I be pre-moving into day folders and only processing a day at a time or something?
I cannot run the CloudFront task without getting this responses.
ERROR TaskSetManager: Total size of serialized results of 3055 tasks (1052.9 MB) is bigger than spark.driver.maxResultSize (1024.0 MB)