From f6d4cb256c1d4e8c1ff3c5bf43d34565173ea9a4 Mon Sep 17 00:00:00 2001
From: Jake Smith <jake.smith@lexisnexisrisk.com>
Date: Fri, 29 Sep 2023 14:59:24 +0100
Subject: [PATCH] HPCC-30413 Add option to always capture post-mortem info

Enabled either globally (i.e. for all components), or for
individual components (e.g. Thor).
Usage:

expert:
  alwaysPostMortem: true

Signed-off-by: Jake Smith <jake.smith@lexisnexisrisk.com>
---
 helm/hpcc/templates/_helpers.tpl | 17 ++++++++++++-----
 initfiles/bin/check_executes     |  5 ++++-
 2 files changed, 16 insertions(+), 6 deletions(-)

diff --git a/helm/hpcc/templates/_helpers.tpl b/helm/hpcc/templates/_helpers.tpl
index cb2d8e1a78c..eed1336e2b7 100644
--- a/helm/hpcc/templates/_helpers.tpl
+++ b/helm/hpcc/templates/_helpers.tpl
@@ -689,13 +689,13 @@ Check that the storage and spill planes for a component exist
 Add command for a component
 */}}
 {{- define "hpcc.componentCommand" -}}
-{{- if .me.valgrind -}}
+ {{- if .me.valgrind -}}
 valgrind
-{{- else if (include "hpcc.hasPlaneForCategory" (dict "root" .root "category" "debug")) -}}
+ {{- else if (include "hpcc.hasPlaneForCategory" (dict "root" .root "category" "debug")) -}}
 check_executes
-{{- else -}}
+ {{- else -}}
 {{ .process }}
-{{- end }}
+ {{- end }}
 {{- end -}}
 
 {{/*
@@ -713,6 +713,9 @@ Add extra args for a component
  {{- $debugPlane := .me.debugPlane | default (include "hpcc.getFirstPlaneForCategory"  (dict "root" .root "category" "debug")) -}}
  {{- include "hpcc.checkPlaneExists" (dict "root" .root "planeName" $debugPlane) -}}
  {{- $prefix := include "hpcc.getPlanePrefix" (dict "root" .root "planeName" $debugPlane) -}}
+  {{- if or (and (hasKey .me "expert") .me.expert.alwaysPostMortem) (and (hasKey .root.Values.global "expert") .root.Values.global.expert.alwaysPostMortem) -}}
+"-a",{{ "\n" }}
+  {{- end -}}
 "-d", {{ $prefix }},
 "--",
 {{ .process | quote }},
@@ -1670,7 +1673,11 @@ args:
  {{- $debugPlane := .me.debugPlane | default (include "hpcc.getFirstPlaneForCategory"  (dict "root" .root "category" "debug")) -}}
  {{- include "hpcc.checkPlaneExists" (dict "root" .root "planeName" $debugPlane) -}}
  {{- $prefix := include "hpcc.getPlanePrefix" (dict "root" .root "planeName" $debugPlane) -}}
- {{- $_ := set $check_cmd "command" (printf "check_executes -d %s -- %s" $prefix .command) -}}
+ {{- $pmd_always_opt := "" -}}
+ {{- if or (and (hasKey .me "expert") .me.expert.alwaysPostMortem) (and (hasKey .root.Values.global "expert") .root.Values.global.expert.alwaysPostMortem) -}}
+  {{- $pmd_always_opt = "-a " -}}
+ {{- end -}}
+ {{- $_ := set $check_cmd "command" (printf "check_executes %s-d %s -- %s" $pmd_always_opt $prefix .command) -}}
 {{- end }}
 - >-
     {{ $check_cmd.command }};
diff --git a/initfiles/bin/check_executes b/initfiles/bin/check_executes
index 6a470b86053..9299015de8d 100755
--- a/initfiles/bin/check_executes
+++ b/initfiles/bin/check_executes
@@ -11,6 +11,7 @@ PMD_PROGNAME=
 PMD_COPYFILES=()
 PMD_DALISERVER=
 PMD_WORKUNIT=
+PMD_ALWAYS=false
 
 while [ "$#" -gt 0 ]; do
   arg=$1
@@ -27,6 +28,8 @@ while [ "$#" -gt 0 ]; do
       f) shift;
          PMD_COPYFILES+=($1)
          ;;
+      a) PMD_ALWAYS=true
+         ;;
       *) usage
          exit
          ;;
@@ -66,7 +69,7 @@ ${PMD_PROGNAME} --logging.postMortem=1000 "$@"
 
 # If it did not exit cleanly, copy some post-mortem info
 retVal=$?
-if [ $retVal -ne 0 ]; then
+if [ $PMD_ALWAYS = true ] || [ $retVal -ne 0 ]; then
   POST_MORTEM_DIR=${PMD_DIRECTORYBASE}/$(hostname)/$(date -Iseconds)
   mkdir -p ${POST_MORTEM_DIR}
   echo "Post-mortem info gathered in $POST_MORTEM_DIR"