Minimac4
Minimac4 copied to clipboard
statgen
Runtime errors
I get this "free(): invalid next size (normal)" error when using Minimac4 It seems the imputation is finished, and the files are generated but the software trying to free the memory and get this error?
Command line
/share/app/glibc-2.17/lib/ld-2.17.so --library-path /share/app/glibc-2.17/lib:/share/app/gcc-5.2.0/lib64:/share/app/gcc-5.2.0/lib:/share/app/libz/zlib-1.2.11/:/share/app/cmake-3.3.2/lib:/opt/gridengine/lib/lx-amd64::/opt/python/lib ~/software/imputation/minimac4 --refHaps ~/refpanel/G1K_P3_M3VCF_FILES_WITH_ESTIMATES/chr21.1000g.Phase3.v5.With.Parameter.Estimates.m3vcf.gz --haps ~/Eagle_Minimac4/imputation/out/0.40/006/phased.0.40.006.vcf.gz --prefix ~/Eagle_Minimac4/imputation/out/0.40/006/minimac.0.40.006 --cpus 1
Error message：
.........
------------------------------------------------------------------------------
END OF PROGRAM
------------------------------------------------------------------------------
*** Error in `~/software/imputation/minimac4': free(): invalid next size (normal): 0x00002b1871aacdf0 ***
======= Backtrace: =========
/share/app/glibc-2.17/lib/libc.so.6(+0x78466)[0x2b17f850f466]
/share/app/glibc-2.17/lib/libc.so.6(+0x7a3b3)[0x2b17f85113b3]
~/software/imputation/minimac4(_ZN11MarkovModelD1Ev+0xb5)[0x423265]
~/software/imputation/minimac4(_ZN8Analysis11RunAnalysisER6StringS1_S1_S1_+0x8bf)[0x42054f]
~/software/imputation/minimac4(_ZN8Analysis17AnalyzeExperimentER6StringS1_S1_S1_R11AllVariable+0x972)[0x421042]
~/software/imputation/minimac4(main+0x919)[0x418239]
/share/app/glibc-2.17/lib/libc.so.6(__libc_start_main+0xfd)[0x2b17f84b684d]
~/software/imputation/minimac4[0x418b7f]
======= Memory map: ========
00400000-0049f000 r-xp 00000000 73b:8ddf2 144116098088930977 ~/software/imputation/minimac4
..........
I ran a series of imputation jobs with different number of samples (from 1 to 42, we are planning to do more later) in combination with other experiment parameters, and I got different errors.
Besides the error mentioned above, the other error types are listed here：
Appending chunk to final output VCF File : SOME_PATH/minimac.0.95.026.dose.vcf.gz
*** Error in `minimac4': munmap_chunk(): invalid pointer: 0x00002ac0381e1f80 ***
Saving Samples in temporary VCF file ...
*** Error in `minimac4': malloc(): smallbin double linked list corrupted: 0x00002ac8cde1c850 ***
Saving Samples in temporary VCF file ...
*** Error in `minimac4': corrupted double-linked list: 0x00002b5bf7148f40 ***
------------------------------------------------------------------------------
END OF PROGRAM
------------------------------------------------------------------------------
*** Error in `minimac4': double free or corruption (!prev): 0x00002ba5c224cde0 ***
I am not sure if all of them were due to the 1.0.0 version I used. Could Anyone send me the executables from version 1.0.1？
This is basic memory corruption. I'm assuming this is an issue with the dynamic loading of a specific version of libc. In any case, it will not be easily debugged via an issues thread. Have you tried building from source using the instructions at https://github.com/statgen/Minimac4/issues/14#issuecomment-441264355
I tried to compile under a Ubuntu 14.04 vm on a cloud server, and copy the executable to our cluster, and use glibc 2.17 to run it. For this v1.0.1 version, 90% of the jobs are finished with out error. but 10% of them still got follow error.
Starting Imputation ...
Imputing Samples 1-37 [100%] out of 37 samples ...
minimac4: /opt/bin/Minimac4/src/MarkovModel.cpp:1102: void MarkovModel::unfoldProbabilities(int, std::vector<float>&, std::vector<float>&, std::vector<float>&, int, std::vector<ReducedHaplotypeInfo>&, int): Assertion `adj_norec[i]<1e18' failed.
The error can be repeated on certain jobs, even including some jobs that have succeed with v1.0.0.
This seems like some kind of precision overflow issue. As of now I would suggest you to use minimac3 instead. In the meantime if you could email me the log files, I can look into it and Try to fix the bug.
On Thu, Dec 6, 2018, 7:55 PM yutoubb <[email protected] wrote:
I tried to compile under a Ubuntu 14.04 vm on a cloud server, and copy the executable to out cluster, and use glibc to run it. For this v1.0.1 version, 90% of the jobs are finished with out error. but 10% of them still got follow error.
Starting Imputation ...
Imputing Samples 1-37 [100%] out of 37 samples ... minimac4: /opt/bin/Minimac4/src/MarkovModel.cpp:1102: void MarkovModel::unfoldProbabilities(int, std::vector
&, std::vector &, std::vector &, int, std::vector<ReducedHaplotypeInfo>&, int): Assertion `adj_norec[i]<1e18' failed. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/statgen/Minimac4/issues/16#issuecomment-445114893, or mute the thread https://github.com/notifications/unsubscribe-auth/AHuICK8LiR2qbB8gNvs41ysTeoq7L-ZQks5u2eawgaJpZM4Y-TER .
I tried to send you an email but got an unknown error. Here is the stderr and stdout: stderr
/opt/gridengine/default/spool/compute-f14-16/job_scripts/7827469: line 1: 194704 Aborted (core dumped) /share/app/glibc-2.17/lib/ld-2.17.so --library-path /share/app/glibc-2.17/lib:/share/app/gcc-5.2.0/lib64:/share/app/gcc-5.2.0/lib:/share/app/libz/zlib-1.2.11/:/share/app/cmake-3.3.2/lib:/opt/gridengine/lib/lx-amd64:/opt/openmpi/lib:/opt/python/lib /software/imputation/minimac4 --refHaps /refpanel/G1K_P3_M3VCF_FILES_WITH_ESTIMATES/chr21.1000g.Phase3.v5.With.Parameter.Estimates.m3vcf.gz --haps /workdir/Eagle_Minimac4/imputation/out/0.10/038/phased.0.10.038.vcf.gz --prefix /workdir/Eagle_Minimac4/imputation/out/0.10/038/minimac.0.10.038 --cpus 1
real 9m41.060s
user 9m14.829s
sys 0m12.227s
stdout
--------------------------------------------------------------------------------
Minimac4 - Fast Imputation Based on State Space Reduction HMM
--------------------------------------------------------------------------------
(c) 2014 - Sayantan Das, Christian Fuchsberger, David Hinds
Mary Kate Wing, Goncalo Abecasis
Version: 1.0.1;
Built: Thu Dec 6 09:08:54 UTC 2018 by root
Command Line Options:
Reference Haplotypes : --refHaps [/refpanel/G1K_P3_M3VCF_FILES_WITH_ESTIMATES/chr21.1000g.Phase3.v5.With.Parameter.Estimates.m3vcf.gz],
--passOnly, --rsid, --referenceEstimates [ON],
--mapFile [docs/geneticMapFile.b38.map.txt.gz]
Target Haplotypes : --haps [/workdir/Eagle_Minimac4/imputation/out/0.10/038/phased.0.10.038.vcf.gz]
Output Parameters : --prefix [/workdir/Eagle_Minimac4/imputation/out/0.10/038/minimac.0.10.038],
--estimate, --nobgzip, --vcfBuffer [200],
--format [GT,DS], --allTypedSites, --meta,
--memUsage
Chunking Parameters : --ChunkLengthMb [20.00], --ChunkOverlapMb [3.00]
Subset Parameters : --chr [], --start, --end, --window
Approximation Parameters : --minimac3, --probThreshold [0.01],
--diffThreshold [0.01], --topThreshold [0.01]
Other Parameters : --log, --help, --cpus [1], --params
PhoneHome : --noPhoneHome, --phoneHomeThinning [50]
URL = http://genome.sph.umich.edu/wiki/Minimac4
Starting Main Imputation/Estimation Analysis ...
Performing preliminary check on input parameters...
------------------------------------------------------------------------------
PRELIMINARY FILE CHECK
------------------------------------------------------------------------------
Checking GWAS haplotype file : /workdir/Eagle_Minimac4/imputation/out/0.10/038/phased.0.10.038.vcf.gz
Gathering variant information ...
Successful !!!
Checking Reference haplotype file : /refpanel/G1K_P3_M3VCF_FILES_WITH_ESTIMATES/chr21.1000g.Phase3.v5.With.Parameter.Estimates.m3vcf.gz
Gathering variant information ...
Successful !!!
Reference Panel : Found 2504 samples (5008 haplotypes) and 653791 variants ...
Target/GWAS Panel : Found 38 samples (76 haplotypes) and 57426 variants ...
57426 variants overlap with Reference panel
0 variants imported that exist only in Target/GWAS panel
------------------------------------------------------------------------------
CHUNKING INFORMATION
------------------------------------------------------------------------------
Chunking region into 1 chunk(s) with atleast 653791 variants in each chunk ...
Details of chunks is given below ...
No LeftBuffer LeftEnd RightPoint RightBuffer #Sites(GWAS/Ref/%)
-------------------------------------------------------------------------------
1 9411245 9411245 48119740 48119740 57426/ 653791/ 8.78%
------------------------------------------------------------------------------
MAIN IMPUTATION ANALYSIS
------------------------------------------------------------------------------
Starting imputation analysis of 1 chunk(s) ...
-------------------------------------------
Analyzing Chunk 1/1 [chr21:9411245-48119740]
-------------------------------------------
Reading chunk from reference panel ...
Reading chunk from target/GWAS panel ...
Compressing reference panel at GWAS sites ...
Re-compression successful (211 seconds) !!!
Starting Imputation ...
Imputing Samples 1-38 [100%] out of 38 samples ...
minimac4: /opt/bin/Minimac4/src/MarkovModel.cpp:1102: void MarkovModel::unfoldProbabilities(int, std::vector<float>&, std::vector<float>&, std::vector<float>&, int, std::vector<ReducedHaplotypeInfo>&, int): Assertion `adj_norec[i]<1e18' failed.
Minimac3 dose not sopport hg19 reference? should I change the chromosome names in vcf and refpanel from chr21 to 21 ?
Checking variant information ...
Error !!! Target VCF File contains chromosome : chr21
VCF File can only contain chromosomes 1-22 and X !!!
Program Aborting ...
Program Exiting ...
No you don’t have to change your reference panel.
Simply add the following option to your command line would work:
--myChromosome chr21
This is a latent option which is not included in the documentation..
Was the source of this error ever found or corrected? We are still encountering this bug in minimac4:
minimac4: /build/minimac4/build/Minimac4/Minimac4/src/MarkovModel.cpp:1085: void MarkovModel::unfoldProbabilities(int, std::vector<float>&, std::vector<float>&, std::vector<float>&, int, std::vector<ReducedHaplotypeInfo>&, int): Assertion 'adj_norec[i]<1e18' failed.
I'd also be interested if the issues described above where resolved. I have a similar runtime error with minimac4:
minimac4: /home/ddidomizio/tmp_src/Minimac4/src/MarkovModel.cpp:755: void MarkovModel::CreatePRefPAlt(ReducedHaplotypeInfo&, int, int): Assertion 'FoldedProbValue[MappedIndex]>=0.0f' failed.
I am attaching my logfile. m4_chr20_chunk2_imputed.txt
Just in case you wonder about my chunking parameters, my input data is already chunked (from my previous pipeline with minimac3) so I chose these parameters to effectively disable the chunking.
@Santy-8128 , any advise? Thank you!
I have the same problem as @yutoubb and @mikesivley:
Minimac4: MarkovModel.cpp:1085: void MarkovModel::unfoldProbabilities(int, std::vector<float>&,
std::vector<float>&, std::vector<float>&, int, std::vector<ReducedHaplotypeInfo>&, int):
Assertion `adj_norec[i]<1e18' failed.
Please let me know if you need access to the log file or dataset.
Please pull the latest from the master branch, delete the contents of your build directory, configure, and build.
@jonathonl I already tried a version based on the latest commit on the master branch after seeing a bunch of probability-related fixes after 1.0.2 was released. Same behavior.
If you have the latest from master, then you need to reconfigure from a clean (e.g., empty) build directory before rebuilding the latest. You would not be getting that message with a clean build from master unless you explicitly enabled a debug build.
You got me there: I was compiling the few .cpp files manually because I didn't want to mess around with CMake / cget and didn't no-op the assert calls in the process. I'm definitely on a fresh checkout on the latest commit, though. If I set -DNDEBUG (as the Release build type would do), the error disappears of course -- but the question is should it and should its presence in the debug build be ignored?
That assertion can be ignored. The 1e18 threshold is too stringent. If those values actually get too large, the results file will be filled with NaN's.