Open HugoJH opened 3 years ago
Hi @HugoJH, I checked out the code and I cannot find the open statement that causes this. Can you give us some more information?
@almahmoud, you've experienced this behavior as well, right?
Yes! The problem seems to be in tool-util I think, cause it came up when running the tool tests directly not just with ephemeris. A temporary workaround was setting ulimit -n 20000
(or some other really high number depending on how many tests are run total)
Are these files or sockets that are open ? Can you also check what version of galaxy-tool-util you've got in your environment ? (pip freeze| grep galaxy-tool-util
)
The last time I tried it was installed from source from https://github.com/galaxyproject/galaxy/pull/10767/files. And as far as I can tell, it's at least files, not sure if sockets as well:
Small snippet of lsof 2>/dev/null | grep ubuntu | sort
(from an older run with ephemeris)
shed-tool 11126 11143 ubuntu 1367u REG 8,1 47 34440 /tmp/tmpzvhs83bdmultiCovBed_result1.bed (deleted)
shed-tool 11126 11143 ubuntu 1368u REG 8,1 259 34448 /tmp/tmp750h1vdqnucBed_result2.bed (deleted)
shed-tool 11126 11143 ubuntu 1369u REG 8,1 47 34442 /tmp/tmpsglw1g2xmultiCovBed_result1.bed
shed-tool 11126 11143 ubuntu 1370u REG 8,1 161 34446 /tmp/tmpfotfnew7nucBed_result1.bed
shed-tool 11126 11143 ubuntu 1371u REG 8,1 259 34450 /tmp/tmpip5lan0gnucBed_result2.bed
shed-tool 11126 11143 ubuntu 1372u REG 8,1 280 34452 /tmp/tmp7b58k0nhnucBed_result3.bed (deleted)
shed-tool 11126 11143 ubuntu 1373u REG 8,1 280 34454 /tmp/tmpn5kbmbx6nucBed_result3.bed
shed-tool 11126 11143 ubuntu 1374u REG 8,1 26 34460 /tmp/tmp7quq5qiisubtractBed_result2.bed (deleted)
shed-tool 11126 11143 ubuntu 1375u REG 8,1 13 34464 /tmp/tmpy_xh9zrksubtractBed_result3.bed (deleted)
shed-tool 11126 11143 ubuntu 1376u REG 8,1 26 34462 /tmp/tmph4t2pdorsubtractBed_result2.bed
shed-tool 11126 11143 ubuntu 1377u REG 8,1 412 34468 /tmp/tmp9c8u2cipbamToBed_result1.bed (deleted)
shed-tool 11126 11143 ubuntu 1378u REG 8,1 26 34458 /tmp/tmpxj_5dtzusubtractBed_result1.bed
shed-tool 11126 11143 ubuntu 1379u REG 8,1 412 34470 /tmp/tmpu4s4z_grbamToBed_result1.bed
Alright, I've traced this too, and https://github.com/galaxyproject/galaxy/pull/10852 prevents accumulating those in a small test.
While doing
shed-tools test
over my galaxy instance, had a problems of reaching the limit of maximum number of open files. For the sake of keep running the test, I modified the limit editing /etc/security/limits. Relaunched theshed-tools test
command and waited enough time to check the number of open files at that point.I did
lsof | grep shed-tool | grep -v deleted| wc
and found out that there where more than 3000 opened files by the shed-tools process. Seems like the files opened for each test are not gettting closed once the test finishes. There should be a mechanism to close all unnecesary file descriptors once the test finishes.