Trinity Chrysalis/Butterfly Out of memory but still running?

Question

[mparker21]

Hi,

I am new to using trinity (and also have very little experience in cs), so I am not too familiar with it, but have been running a de novo assembly on some RNAseq reads using the following command:

Trinity --seqType fq --max_memory 230G --monitoring --SS_lib_type RF --CPU 64 --output [output path] --trimmomatic --verbose --min_kmer_cov 2 --left [left path] --right [right path]

I originally ran the program at --max_memory 250GB (server has 256GB), but the program killed during the inchworm stage. The person running the server (that I am using to run this) told me to decrease the parameter to 230GB, which worked. Now the program gets through the first few stages fine, but now that it is at the Chryaslis/Butterfly stage it is running out of memory. The program keeps throwing out of memory errors, but its also still running and the % completed is increasing (albeit slowly). Should I just let it keep running?

here is memory error I have been getting while its still running: Error occurred during initialization of VM java.lang.OutOfMemoryError: unable to create new native thread

I also got this: sh: fork: retry: No child processes] sh: fork: retry: No child processes sh: fork: retry: No child processes sh: fork: retry: No child processes sh: fork: retry: No child processes sh: fork: retry: No child processes sh: fork: retry: No child processes sh: fork: Resource temporarily unavailable Error, cmd: /usr/local/Trinity-v2.4.0/util/support_scripts/filter_iworm_by_min_length_or_cov.pl

followed by Trinity run failed. Must investigate error above.

thanks for the help. I also don't have access to a higher memory server so if anyone has any suggestions on what to do that would be appreciated.

[Michael]

Hi, see here: Trinity: how to avoid out of memory error in butterfly

The only trick that worked for me is to run Trinity twice, but let the first run finish despite errors, in the second run reduce --CPU to 4,8, or 1, then it will pick up and re-run the failed jobs from the first run, and as it makes fewer threads, it will use less memory. Repeat until it finishes.

[mparker21]

thanks! I will try this