MMseqs2 Release 18 Error in fwbw Alignment (linclust function, Uniref50 dataset)

[JaceWA99]

Hello, I am deploying MMseqs on Databricks for data pipelining work, and ran into an issue while testing the "linclust" function (separated out easy-linclust for testing). The test dataset is Uniref50 [https://www.uniprot.org/uniref?query=%28identity%3A0.5%29].

When running on the current release (cf9e9f2), clustering failed on the fwbw alignment step (shortened logging below).

I then pinned to release version 17 (b804fbe) and the function performed as normal, indicating that there is an issue with the new fwbw alignment step when processing Uniref50.

For now I can continue using a pinned version, but this would be good to resolve in the future.

NOTE: After a number of tries, the current version succeeded a couple times, indicating a random error. Majority of tests hit the alignment error.

Code (personal paths removed)
mmseqs createdb uniref50.fasta uniref50_db
mmseqs linclust uniref50_db uniref50_clusters tmpA --min-seq-id 0.5 -c 0.5 --similarity-type 2 --cov-mode 1
mmseqs createtsv uniref50_db uniref50_db uniref50_clusters uniref50_clusters.tsv

Logs (config and error)
MMseqs Version:       cf9e9f2605b15436eea129dc42961ec411e57af4
Cluster mode       0
Max connected component depth       1000
Similarity type       2
Threads       32
Compressed       0
Verbosity       3
Weight file name       
Cluster Weight threshold       0.9
Set mode       false
Substitution matrix       aa:blosum62.out,nucl:nucleotide.out
Add backtrace       false
Alignment mode       2
Alignment mode       0
Allow wrapped scoring       false
E-value threshold       0.001
Seq. id. threshold       0.5
Min alignment length       0
Seq. id. mode       0
Alternative alignments       0
Coverage threshold       0.5
Coverage mode       1
Max sequence length       65535
Compositional bias       1
Compositional bias scale       1
Max reject       2147483647
Max accept       2147483647
Include identical seq. id.       false
Preload mode       0
Pseudo count a       substitution:1.100,context:1.400
Pseudo count b       substitution:4.100,context:5.800
Score bias       0
Realign hits       false
Realign score bias       -0.2
Realign max seqs       2147483647
Correlation score weight       0
Gap open cost       aa:11,nucl:5
Gap extension cost       aa:1,nucl:2
Zdrop       40
Alphabet size       aa:21,nucl:5
k-mers per sequence       21
Spaced k-mers       0
Spaced k-mer pattern       
Scale k-mers per sequence       aa:0.000,nucl:0.200
Adjust k-mer length       false
Mask residues       0
Mask residues probability       0.9
Mask lower case residues       0
Mask lower letter repeating N times       0
k-mer length       0
Shift hash       67
Split memory limit       0
Include only extendable       false
Skip repeating k-mers       false
Rescore mode       0
Remove hits by seq. id. and coverage      false
Sort results       0
Remove temporary files       false
Force restart with latest tmp       false
MPI runner       

Compute score and coverage
Query database size: 48398278 type: Aminoacid
Target database size: 48398278 type: Aminoacid
Calculation of alignments
r.word: 2
bests_reverse.first.score: 14565
r.score1: 14563
Score of forward/backward SW differ. This should not happen.
Start: Q: 16052, T: 14692. End: Q: 19494, T 18103
[===============================Error: Alignment step died