salmon: alignment group queue pool has been exhausted

[Joel Parker]

When running on multiple rat rna-seq libraries with Salmon we encountered a message that we have not observed in runs from other organisms:

 

processed 3000000 reads in current round[2016-03-03 16:06:00.271] [jointLog] [info]

 

The alignment group queue pool has been exhausted.  722834 extra fragments were allocated on the heap to saturate the pool.  No new fragments will be allocated

Salmon is still producing the quant.sf output and it looks like normal output, but we are a bit concerned that quantification has been affected in some way.

 

Any input would be appreciated!

[Rob]

Hi Joel,

  This message is completely normal, and has no effect on the quantification results.  Basically, since Salmon is multithreaded, there can be a difference between the rate it which it can read alignments from the BAM file, and the rate it which it can process those alignments for quantification.  Salmon maintains a "buffer" of reads that have already been read from file, but which have not been processed.  However, to prevent the memory usage of Salmon from growing too large, this buffer has a fixed maximum size.  This is the message that is printed when the buffer is reaches its maximum size.  All that it means is that, instead of allocating more memory, it will wait for some of those alignments to be processed before reading more from file.  However, all of the alignments will still be read and considered for purposes of quantification.  I hope this clears things up.  Let me know if you have any other questions.

Best,

Rob

[Joel Parker]

Great, thanks!

Also, probably no surprise to salmon users, but we are very pleased with this method.  We find extraordinarily high concordance in relative expression estimates between Mapsplice/RSEM and STAR/Salmon - far superior to other methods and combinations that we attempted, and enormous computational benefit with the update.

[Rob]

Awesome; thanks for the positive feedback!

--Rob