(travis) Better diagnostics on *why* a build failed, reduce parellelism

This commit resulted in https://github.com/travis-ci/travis-ci/issues/2904

To mitigate run everything under 10 (not 32, not even 16) threads, hope
things will be allright. Although the real fix would be to 1) get the above
mentioned issue on travis fixed and 2) tuning the rdbms daemons to eat less
memory (both of these options are problematic).

The diags are *indispensable* when travis wakes up the OOMKiller
As a bonus add getstatus to all subshells for even better idea wtf just went
wrong

diff --git a/.travis.yml b/.travis.yml
index cc065f1..d68a13d 100644 (file)
--- a/.travis.yml
+++ b/.travis.yml
@@ -254,24 +254,27 @@ install:
 before_script:
   # Preinstall/install deps based on envvars/CLEANTEST
   #
-  - maint/travis-ci_scripts/30_before_script.bash
+  # need to invoke the after_failure script manually
+  # because 'after_failure' runs only after 'script' fails
+  #
+  - maint/getstatus maint/travis-ci_scripts/30_before_script.bash || ( maint/travis-ci_scripts/50_after_failure.bash && /bin/false )
 
 script:
   # Run actual tests
   #
-  - maint/travis-ci_scripts/40_script.bash
+  - maint/getstatus maint/travis-ci_scripts/40_script.bash
 
 after_success:
   # Check if we can assemble a dist properly if not in CLEANTEST
   #
-  - maint/travis-ci_scripts/50_after_success.bash
+  - maint/getstatus maint/travis-ci_scripts/50_after_success.bash
 
 after_failure:
-  # No tasks yet
+  # Final sysinfo printout on fail
   #
-  #- maint/travis-ci_scripts/50_after_failure.bash
+  - maint/getstatus maint/travis-ci_scripts/50_after_failure.bash
 
 after_script:
   # No tasks yet
   #
-  #- maint/travis-ci_scripts/60_after_script.bash
+  #- maint/getstatus maint/travis-ci_scripts/60_after_script.bash

diff --git a/maint/travis-ci_scripts/10_before_install.bash b/maint/travis-ci_scripts/10_before_install.bash
index ddc2589..da10ee4 100755 (executable)
--- a/maint/travis-ci_scripts/10_before_install.bash
+++ b/maint/travis-ci_scripts/10_before_install.bash
@@ -38,10 +38,9 @@ if [[ -n "$SHORT_CIRCUIT_SMOKE" ]] ; then return ; fi
 # slurp the entire file and get the index off the last
 # `processor    : XX` line
 #
-# We also divide the result by two, otherwise the travis VM
-# gets overloaded (the amount of available swap is just TOOOO
-# damn small)
-export NUMTHREADS="$(( ( $(perl -0777 -n -e 'print (/ (?: .+ ^ processor \s+ : \s+ (\d+) ) (?! ^ processor ) /smx)' < /proc/cpuinfo) + 1 ) / 2 ))"
+# We also divide the result by a factor, otherwise the travis VM gets
+# overloaded (the amount of available swap is just TOOOO damn small)
+export NUMTHREADS="$(( ( $(perl -0777 -n -e 'print (/ (?: .+ ^ processor \s+ : \s+ (\d+) ) (?! ^ processor ) /smx)' < /proc/cpuinfo) + 1 ) / 3 ))"
 
 export CACHE_DIR="/tmp/poormanscache"
 

diff --git a/maint/travis-ci_scripts/50_after_failure.bash b/maint/travis-ci_scripts/50_after_failure.bash
index fb5b5fb..f087d6f 100755 (executable)
--- a/maint/travis-ci_scripts/50_after_failure.bash
+++ b/maint/travis-ci_scripts/50_after_failure.bash
@@ -1,11 +1,12 @@
 #!/bin/bash
 
-# !!! Nothing here will be executed !!!
-# The source-line calling this script is commented out in .travis.yml
-
 # this file is executed in a subshell - set up the common stuff
 source maint/travis-ci_scripts/common.bash
 
 if [[ -n "$SHORT_CIRCUIT_SMOKE" ]] ; then exit 0 ; fi
 
-echo_err "Nothing to do"
+echo_err "
+$(ci_vm_state_text)
+
+=== dmesg ringbuffer
+$(sudo dmesg)"