fatal error

[guptapa]

Hi,
I am using STAR-2.6.0a for custom genome index and paired end reads and getting following error:

EXITING because of fatal error: buffer size for SJ output is too small
Solution: increase input parameter --limitOutSJcollapsed
Aug 23 08:37:22 ...... FATAL ERROR, exiting

options I used:
--genomeLoad NoSharedMemory --genomeDir --runThread 1 --readFilesIn --outSAMtype BAM Unsorted --limitOutSJcollapsed int>0 --outFileNamePrefix

same error with and without using this flag : --limitOutSJcollapsed int>0

[alexdobin]

Hi Paul,

--limitOutSJcollapsed requires a positive integer number. The default value is usually fine, please first try without specifying this parameter at all.
Also, I would recommend using more recent stable releases such as 2.6.1d or 2.7.2a .

Cheers
Alex

[guptapa]

Hi Alex, As per your recommendation I used 2.7.2a but again got same error: EXITING because of fatal error: buffer size for SJ output is too small Solution: increase input parameter --limitOutSJcollapsed Aug 24 00:06:49 ...... FATAL ERROR, exiting I didn’t use —limitOutSJcollapsed flag in my script. Parul

…

On Aug 23, 2019, at 10:59 AM, Alexander Dobin ***@***.***> wrote: Hi Paul, --limitOutSJcollapsed requires a positive integer number. The default value is usually fine, please first try without specifying this parameter at all. Also, I would recommend using more recent stable releases such as 2.6.1d or 2.7.2a . Cheers Alex — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#719?email_source=notifications&email_token=ADSLCGOSTCOLZ3CCXNT7LGDQGACOVA5CNFSM4IPBBNMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5AT4VY#issuecomment-524369495>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADSLCGM3HJLL5E5SNPRC2N3QGACOVANCNFSM4IPBBNMA>.

[alexdobin]

Hi Parul,

please send me the Log.out file of this failed run.

Cheers
Alex

[guptapa]

Here you go! Parul

…

On Aug 26, 2019, at 11:06 AM, Alexander Dobin ***@***.***> wrote: Hi Parul, please send me the Log.out file of this failed run. Cheers Alex — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#719?email_source=notifications&email_token=ADSLCGIYNMJH6VHMNRCD2TDQGP5RBA5CNFSM4IPBBNMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5E2TMI#issuecomment-524921265>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADSLCGJPNIYYH47EOMRGOOTQGP5RBANCNFSM4IPBBNMA>.

[alexdobin]

Hi Parul,

it did not get attached. I think you need to do it from the web-site.

Cheers
Alex

[guptapa]

Hi Alex, I hope this would work, let me know if not. Thanks, Parul

…

On Aug 26, 2019, at 11:32 AM, Alexander Dobin ***@***.***> wrote: Hi Parul, it did not get attached. I think you need to do it from the web-site. Cheers Alex — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#719?email_source=notifications&email_token=ADSLCGOZM6SORFDFPP274LTQGQAQVA5CNFSM4IPBBNMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5E42RY#issuecomment-524930375>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ADSLCGPJS3PAPTNF3OTM5UDQGQAQVANCNFSM4IPBBNMA>.

[alexdobin]

No, still no attachment... You can email it to me: dobin @ cshl . edu

[alexdobin]

Got it, thanks!
Nothing suspicious in the Log.out file.
So it seems like you do need to increase --limitOutSJcollapsed .
The default value is 1000000, I would try increasing it to 2000000 and if it does not help, 5000000, i.e.
--limitOutSJcollapsed 2000000

Cheers
Alex

[guptapa]

Thanks much Alex. Parul

…

On Aug 26, 2019, at 12:52 PM, Alexander Dobin ***@***.***> wrote: --limitOutSJcollapsed 2000000

[DiracZhu1998]

Got it, thanks!
Nothing suspicious in the Log.out file.
So it seems like you do need to increase --limitOutSJcollapsed .
The default value is 1000000, I would try increasing it to 2000000 and if it does not help, 5000000, i.e.
--limitOutSJcollapsed 2000000

Cheers
Alex

Hi Alex, I couldn't figure out the explanation of this parameter < limitOutSJcollapsed >, may you specify it again? Thanks a lot !
Dirac

[alexdobin]

Hi Dirac,

STAR records the splice junctions detected in the mapped reads, and --limitOutSJcollapsed defines the maximum number of splice junctions. It is used to pre-allocate arrays to store SJ information.

Cheers
Alex

[tallulandrews]

I get the same error working with some single nucleus RNAseq data. Is there a parameter somewhere I should change for single nucleus RNAseq data (lots of intronic / immature mRNAs), or perhaps I should edit the gtf file?

[alexdobin]

Hi @tallulandrews

if you get the same error, you need to increase --limitOutSJcollapsed from the default 1000000 to 2000000 or even larger.
Generally, there are no specific parameters for nuclear RNA.

Cheers
Alex

[alanhoyle]

Is tweaking --limitOutSJcollapsed always the diagnosis with exit code 104?

[alexdobin]

Hi Alan,

yes, this problem will always generate exit code 104, but other problems may generate this code as well.

[alanhoyle]

We're trying to make a workflow that auto-scales resources when a job fails. Is the only way to determine what to scale to parse the logs?

The current version starts by explictly setting the default: --limitOutSJcollapsed 100000 and doubles on each attempt (up to 8000000)

Is there a list of exit codes? Is there a list of things that would scale automatically? Historically, this is the only error we've seen that gets addressed this way, but I haven't reviewed the code to see if there are others.

Thank you @alexdobin !

[alexdobin]

Hi Alan,

in the beginning, I was trying to keep different exit codes for different classes of errors, but since no one was using them, I stopped doing carefully.
I think I should just make automatically reallocate arrays that are controlled by this parameter to avoid it altogether.
If there are other similar situations, please let me know. Most of them should be triggered very rarely as long as you close to default parameters, and the samples are not very extreme.