From 73e2eea4be12ac94b38c55359195799ba2c4cf17 Mon Sep 17 00:00:00 2001
From: Olamide Isreal <holarmidhey07@gmail.com>
Date: Tue, 28 Jul 2026 10:05:17 +0100
Subject: [PATCH] Stop staging the 9.2 GB ChEMBL DB into every task workdir

GET_FINAL_METABOLITES_STATIC took chembl_db as a `path` input, so Nextflow
copied the ~9.2 GB SQLite DB into each task's work directory. With ~29
concurrent tasks that is hundreds of GB of I/O, against a process that also
declared only 1 GB of memory. Every task failed with exit 1 and retried ten
times, producing 746 errors and an empty 1b_final_metabolites/ output.

Pass the DB as a `val` path instead so tasks read it in place from the
dreamdock-data PVC, and give the process parameterised growing memory.
---
 main.nf                | 13 ++++++++++++-
 main_biotransformer.nf | 13 +++++++++----
 2 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/main.nf b/main.nf
index c11dd9c..9eca350 100644
--- a/main.nf
+++ b/main.nf
@@ -29,6 +29,14 @@ params.bt_max_retries = 10
 params.bt_fail_action = 'ignore' // 'terminate' or 'ignore'
 params.bt_max_forks = 0             // 0 = unlimited, set to N to limit concurrency
 
+// GET_FINAL_METABOLITES_STATIC — queries the ~9.2 GB ChEMBL SQLite DB on the PVC.
+// The DB is opened read-only/immutable and is NOT staged into the task workdir, so the
+// memory below covers rdkit + the biotransformer CSV, not the database itself.
+params.chembl_initial_memory = 4         // GB - starting memory
+params.chembl_growth_memory  = 4         // GB - additional memory per retry attempt
+params.chembl_max_retries    = 3
+params.chembl_fail_action    = 'ignore'  // 'terminate' or 'ignore' after retries exhausted
+
 //CONPLEX
 params.keep_enst = 'false' //'true' //'false' //to keep individual protein data created by conplex step
 params.conplex_initial_memory = 5   // GB - starting memory for conplex
@@ -121,7 +129,10 @@ workflow {
         default:
             println("Invalid mode specified: ${params.mode}")
     }
-    chembl_ch = Channel.fromPath(params.chembl_db).collect()
+    // Pass the ChEMBL DB as a plain path string rather than a staged file: at ~9.2 GB,
+    // staging it per task copied the DB into every work directory. The process reads it
+    // read-only from the PVC mount instead.
+    chembl_ch = Channel.value(params.chembl_db)
     // GET_FINAL_METABOLITES(HUMAN_TRANSFORMER.out)    // TODO: LOCALIZE
     GET_FINAL_METABOLITES_STATIC(HUMAN_TRANSFORMER.out, chembl_ch)
     //CONPLEX
diff --git a/main_biotransformer.nf b/main_biotransformer.nf
index 23ec209..c07e28e 100755
--- a/main_biotransformer.nf
+++ b/main_biotransformer.nf
@@ -274,18 +274,23 @@ process GET_FINAL_METABOLITES {
 }
 
 process GET_FINAL_METABOLITES_STATIC {
-    memory 1.GB
+    // The ChEMBL DB is ~9.2 GB. It is passed as a `val` path (not a staged `path`
+    // input) so Nextflow does not copy it into every task work directory — with
+    // concurrent tasks that multiplied into hundreds of GB of I/O. The DB lives on
+    // the dreamdock-data PVC and is opened read-only/immutable, so sharing one copy
+    // across tasks is safe.
+    memory { params.chembl_initial_memory.toInteger().GB + (task.attempt - 1) * params.chembl_growth_memory.toInteger().GB }
 
     container "${params.container_chembl}"
     containerOptions "${params.containerOptions}"
     publishDir "${params.outdir}/${params.project_name}/1b_final_metabolites", mode: 'copy'
 
-    errorStrategy { task.attempt <= 10 ? 'retry' : 'ignore' }
-    maxRetries 10
+    errorStrategy { task.attempt <= params.chembl_max_retries.toInteger() ? 'retry' : params.chembl_fail_action }
+    maxRetries { params.chembl_max_retries.toInteger() }
 
     input:
     path smiles_csv
-    path chembl_db  // Add ChEMBL SQLite database as input
+    val chembl_db  // Path to the ChEMBL SQLite DB on the PVC (not staged into the workdir)
 
     output:
     path "${smiles_csv.simpleName}.txt"