RAxML-NG: ERROR: vector::_M_default_append

[Agnès Baud]

Hi,

A few weeks ago, I ran this command following what was suggested by `raxml-ng --parse --msa 16S_CustomDB_MA_auto.fasta.reduced --model GTR+G` (estimated memory requirements: 44728 MB; recommended number of threads / MPI processes: 8) with a bit more memory just in case:
`sbatch --mem=120G --qos=atm -p atm -c 8 launch_RAxML.sh`

==> launch_RAxML.sh <==
phyMA=16S_CustomDB_MA_auto.fasta.reduced.raxml.reduced.phy
raxml-ng --all --msa $phyMA --model GTR+G --tree pars{10} --bs-trees autoMRE{500} --seed 1234 --threads 8

I let it run for 13 days and then I became worried. I saw that it hadn't written anything in the log since the first day.

==> log slurm <==
(...)
Parallelization scheme autoconfig: 1 worker(s) x 8 thread(s)
(...)
[00:00:08] Generating 10 parsimony starting tree(s) with 19443 taxa
[02:57:11] Data distribution: max. partitions/sites/weight per thread: 1 / 1178 / 18848
[02:57:20] Data distribution: max. searches per worker: 510

Starting ML tree search with 10 distinct starting trees

I canceled the job and tried to run it again, this time with more cpus.
`sbatch --mem=400G --qos=atm -p atm -c 48 launch_RAxML.sh`

==> launch_RAxML.sh <==
raxml-ng --all --msa $phyMA --model GTR+G --tree pars{10} --bs-trees autoMRE{500} --seed 1234 --threads 48

But I got this error in the log:
==> log slurm <==
(...)
WARNING: The model you specified on the command line (GTR+G) will be ignored
         since the binary MSA file already contains a model definition.
         If you want to change the model, please re-run RAxML-NG
         with the original PHYLIP/FASTA alignment and --redo option.
(...)
Parallelization scheme autoconfig: 8 worker(s) x 6 thread(s)
(...)
NOTE: Per-rate scalers were automatically enabled to prevent numerical issues on taxa-rich alignments.
NOTE: You can use --force switch to skip this check and fall back to per-site scalers.

Parallel reduction/worker buffer size: 1 KB  / 0 KB


ERROR: vector::_M_default_append

Can you help me understand what went wrong? Also what is the warning saying since I ran all my commands on the model GTR+G?

Thank you in advance,
Agnès

[Oleksiy Kozlov]

Hi Agnès,

you used a newer raxml-ng version for the 2nd run (1.2.2), which failed to load a checkpoint from a
much older version used for the 1st run (1,0.1). You can fix it by adding the "--redo" flag, which
will ignore the checkpoint and restart the analysis from scratch.

However, please note that tree search and especially bootstrapping will take (very) long time for
such a large dataset (20,000 taxa). If possible, please consider reducing the number of taxa by
subsampling/clustering and/or run a tree search without bootstrapping first.

Alternatively, you can try the developmental version of raxml-ng which can be much faster on such
taxa-rich datasets, and also supports faster-to-compute branch support metrics such as SH-aLRT and
EBG (use --bs-metric option):

https://github.com/amkozlov/raxml-ng/wiki/Installation#building-development-branch

Best,
Oleksiy

On 13.02.25 12:05, Agnès Baud wrote:
> Hi,
>
> A few weeks ago, I ran this command following what was suggested by `raxml-ng --parse --msa
> 16S_CustomDB_MA_auto.fasta.reduced --model GTR+G` (estimated memory requirements: 44728 MB;
> recommended number of threads / MPI processes: 8) with a bit more memory just in case:
> `sbatch --mem=120G --qos=atm -p atm -c 8 launch_RAxML.sh`
>

> *==> launch_RAxML.sh <==*

> phyMA=16S_CustomDB_MA_auto.fasta.reduced.raxml.reduced.phy
> raxml-ng --all --msa $phyMA --model GTR+G --tree pars{10} --bs-trees autoMRE{500} --seed 1234 --
> threads 8
>
> I let it run for 13 days and then I became worried. I saw that it hadn't written anything in the log
> since the first day.
>

> *==> log slurm <==*

> (...)
> Parallelization scheme autoconfig: 1 worker(s) x 8 thread(s)
> (...)
> [00:00:08] Generating 10 parsimony starting tree(s) with 19443 taxa
> [02:57:11] Data distribution: max. partitions/sites/weight per thread: 1 / 1178 / 18848
> [02:57:20] Data distribution: max. searches per worker: 510
>
> Starting ML tree search with 10 distinct starting trees
>
> I canceled the job and tried to run it again, this time with more cpus.
> `sbatch --mem=400G --qos=atm -p atm -c 48 launch_RAxML.sh`
>

> *==> launch_RAxML.sh <==
> *raxml-ng --all --msa $phyMA --model GTR+G --tree pars{10} --bs-trees autoMRE{500} --seed 1234 --

> threads 48
>
> But I got this error in the log:

> *==> log slurm <==
> *(...)

> WARNING: The model you specified on the command line (GTR+G) will be ignored
>          since the binary MSA file already contains a model definition.
>          If you want to change the model, please re-run RAxML-NG
>          with the original PHYLIP/FASTA alignment and --redo option.
> (...)
> Parallelization scheme autoconfig: 8 worker(s) x 6 thread(s)
> (...)
> NOTE: Per-rate scalers were automatically enabled to prevent numerical issues on taxa-rich alignments.
> NOTE: You can use --force switch to skip this check and fall back to per-site scalers.
>
> Parallel reduction/worker buffer size: 1 KB  / 0 KB
>
>
> ERROR: vector::_M_default_append
>
> Can you help me understand what went wrong? Also what is the warning saying since I ran all my
> commands on the model GTR+G?
>
> Thank you in advance,
> Agnès
>

> --
> You received this message because you are subscribed to the Google Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to
> raxml+un...@googlegroups.com <mailto:raxml+un...@googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/raxml/905e799f-2a8e-435c-
> afab-01c72d685822n%40googlegroups.com <https://groups.google.com/d/msgid/raxml/905e799f-2a8e-435c-
> afab-01c72d685822n%40googlegroups.com?utm_medium=email&utm_source=footer>.