Memory for PepQuery2

[mira-miracoli]

Not 100% sure where to place this issue, but I thought it might be interesting for all users of the shared database. Otherwise I can of course move it to EU.

I am currently debugging an error the pepquery2 tool. The job errored because the JVM run out of memory.
When I tried to run the job locally I had to stop at 14G because my laptop (16G) started to lag.
I noticed that:

• it uses 8 cores (from the logs) even though it was allocated 1 core
• the index creation set is much slower on the server than on my laptop (might be storage related)
• there is no rule for TPV (and was no rule for sortinghat)

I would like to change that, but I am not sure which values I should consider. In their documentation I found a recommendation for 8 GB of memory and 4 CPUs which is too little for at least the job I am looking at. When I tried to use gxadmin query tool-memory-per-inputs I found:

    id    |                                tool_id                                 | input_count | total_input_size_mb | mean_input_size_mb | median_input_size_mb | memory_used_mb | memory_used_per_input_mb | memory_mean_input_ratio | memory_median_input_ratio                                                                                                               
----------+------------------------------------------------------------------------+-------------+---------------------+--------------------+----------------------+----------------+--------------------------+-------------------------+--------------------------- 
 ######### | toolshed.g2.bx.psu.edu/repos/galaxyp/pepquery2/pepquery2/2.0.2+galaxy0 |           1 |                  36 |                 36 |                   36 |         283829 |            7948 |                    7948 |                      7948                           

While gxadmin report job-info returned the following:

## Destination Parameters                                                                                                                                                                    
                                                                                 
Key | Value                                                                                                                                                                                  
--- | ---
+Group | `""`
accounting_group_user | `#####`
description | `pepquery2`                                                                                                                                                                    
docker_memory | `3.8G`                                                    
metadata_strategy | `extended`                                                               
request_cpus | `1`                                                                                                                                                                           
request_memory | `3.8G`
requirements | `(GalaxyGroup  ==  "compute")`
submit_request_gpus | `0`    

I am now trying to figure out how to implement a rule here and if we have to change something in the wrapper because of the CPU usage. Since I never used the tool myself I would be happy about any hints from people who have some experience with it.

[mira-miracoli]

I increased the memory for the erroring job to 16G and it finished.
Since I have not enough data to come up with a sensible rule, I would suggest to set the mem to 16G for now

[mira-miracoli]

Unfortunately the tool seems not to be satisfied and condor complains it tries to exceed the 16G:

007 (44662823.000.000) 07/12 09:25:00 Shadow exception!
        Error from slot1_5@vgcnbwc-worker-c36m225-4528.novalocal: Job has gone over memory limit of 16384 megabytes. Peak usage: 16331 megabytes.
        0  -  Run Bytes Sent By Job
        0  -  Run Bytes Received By Job
...
012 (44662823.000.000) 07/12 09:25:00 Job was held.
        Error from slot1_5@vgcnbwc-worker-c36m225-4528.novalocal: Job has gone over memory limit of 16384 megabytes. Peak usage: 16331 megabytes.
        Code 34 Subcode 0

The -Xmx16g was set, so this should not happen. However, since this job seems to need more than 16Gb, we could try to increase it further

[mira-miracoli]

I am not an Java expert, but I would assume, that there is some kind of overhead that adds to the 16GB that the JVM uses as heap

EDIT: This is what I learned so far. Since the wrapper currently defines -Xmx{mem}g, I need to change the wrapper accordingly

[nuwang]

Looks like it supports a -cpu parameter: http://www.pepquery.org/document.html#saparameter
and the wrapper will indeed need to be modified.

[mira-miracoli]

f no one opposes, I would open a PR with that set accordingly in _JAVA_OPTIONS

Looks like it supports a -cpu parameter: http://www.pepquery.org/document.html#saparameter and the wrapper will indeed need to be modified.

yes by default it uses all cores available to it, but it would be cleaner to use it, I guess.

[mira-miracoli]

I got this from Galaxy, probably, because I run je job manually and it was still watched by the Galaxy Handlers:

Job Metrics
cgroup

CPU Time | 2 hours and 54 minutes
-- | --
Failed to allocate memory count | 0E-7
Memory limit on cgroup (MEM) | 48.0 GB
Max memory usage (MEM) | 17.5 GB
Memory limit on cgroup (MEM+SWP) | 8.0 EB
Max memory usage (MEM+SWP) | 17.5 GB
OOM Control enabled | No
Was OOM Killer active? | No
Memory softlimit on cgroup | 0 bytes

...

Destination Parameters

Runner | condor
-- | --
Runner Job ID | 44667975
Handler | handler_sn06_3
+Group | ""
accounting_group_user | 55103
description | pepquery2
docker_memory | 16G
metadata_strategy | extended
request_cpus | 1
request_memory | 16G

[mira-miracoli]

The job is stopped by condor for exceeding its memory:

007 (xxxxxxxxx.000.000) 07/14 11:35:09 Shadow exception!                                                                                                                                      
        Error from slot1_8@vgcnbwc-worker-xxxxxxx: Job has gone over memory limit of 16384 megabytes. Peak usage: 16328 megabytes.                                            
        0  -  Run Bytes Sent By Job                                                                                       
        0  -  Run Bytes Received By Job                                                                                   
...                                                                                                                                                                                          
012 (xxxxxxxxxxxx.000.000) 07/14 11:35:09 Job was held.                                                                                                                                          
        Error from slot1_8@vgcnbwc-worker-xxxxxxxxl: Job has gone over memory limit of 16384 megabytes. Peak usage: 16328 megabytes.                                            
        Code 34 Subcode 0                                                                                                                                                                    
...                                                                                                                                                                                          
013 (xxxxxxxxxx.000.000) 07/14 11:45:02 Job was released.                                                                                                                                      
        via condor_release (by user galaxy)      

Here is a PR to increase it

• increase to 20G #32
  And here for the Wrapper:
• fix PepQuery2 memory galaxyproteomics/tools-galaxyp#722