From e4c886fa09cae5baf0aa3e54793c69139868fa3b Mon Sep 17 00:00:00 2001
From: Jim Garlick <garlick.jim@gmail.com>
Date: Tue, 26 Apr 2022 13:42:58 -0700
Subject: [PATCH] testsuite: cover offline KVS garbage collection

Problem: there is no test coverage for offline KVS garbage
collection.

Add a sharness script that exercises this functionality.
Augment the shutdown-cmd sharness script to cover new shutdown options.
---
 t/Makefile.am                 |  1 +
 t/t2808-shutdown-cmd.t        | 50 +++++++++++++++++++
 t/t2810-kvs-garbage-collect.t | 92 +++++++++++++++++++++++++++++++++++
 3 files changed, 143 insertions(+)
 create mode 100755 t/t2810-kvs-garbage-collect.t

diff --git a/t/Makefile.am b/t/Makefile.am
index e1924b87b771..bfd143047c64 100644
--- a/t/Makefile.am
+++ b/t/Makefile.am
@@ -187,6 +187,7 @@ TESTSCRIPTS = \
 	t2807-dump-cmd.t \
 	t2808-shutdown-cmd.t \
 	t2809-job-purge.t \
+	t2810-kvs-garbage-collect.t \
 	t2900-job-timelimits.t \
 	t3000-mpi-basic.t \
 	t3001-mpi-personalities.t \
diff --git a/t/t2808-shutdown-cmd.t b/t/t2808-shutdown-cmd.t
index 789a21c7e287..fb6f1682cd38 100755
--- a/t/t2808-shutdown-cmd.t
+++ b/t/t2808-shutdown-cmd.t
@@ -136,4 +136,54 @@ test_expect_success 'flux-shutdown as initial program does not hang' '
 	test_expect_code 129 run_timeout 30 flux start flux shutdown
 '
 
+test_expect_success 'submit batch script and wait for it to start' '
+	rm -f job6-has-started &&
+	cat >batch6.sh <<-EOT &&
+	#!/bin/sh
+	flux mini run /bin/true
+	touch job6-has-started
+	sleep 300
+	EOT
+	chmod +x batch6.sh &&
+	flux mini batch -t30m -n1 batch6.sh >jobid6 &&
+	$waitfile job6-has-started
+'
+
+test_expect_success 'one job has run in the batch job' '
+	(FLUX_URI=$(flux uri --local $(cat jobid6)) \
+	    flux jobs -n -a -o {id}) >job6_list &&
+	test $(wc -l <job6_list) -eq 1
+'
+
+test_expect_success 'shutdown batch script with --dump' '
+	(FLUX_URI=$(flux uri --local $(cat jobid6)) \
+	    flux shutdown --dump=dump.tgz)
+'
+test_expect_success 'dump file was created' '
+	tar tvf dump.tgz
+'
+test_expect_success 'restart batch script from dump and wait for it to start' '
+	rm -f job6-has-started &&
+	flux mini batch -t30m -n1 \
+	    --broker-opts=-Scontent.restore=dump.tgz \
+	    batch6.sh >jobid6_try2 &&
+	$waitfile job6-has-started
+'
+test_expect_success 'two jobs have been run in batch job' '
+	(FLUX_URI=$(flux uri --local $(cat jobid6_try2)) \
+	    flux jobs -n -a -o {id}) >job6_list_try2 &&
+	test $(wc -l <job6_list_try2) -eq 2
+'
+test_expect_success 'job id from before restart is in job listing' '
+	grep $(cat job6_list) job6_list_try2
+'
+
+test_expect_success 'shutdown batch script with --gc' '
+	(FLUX_URI=$(flux uri --local $(cat jobid6_try2)) \
+	    flux shutdown --gc)
+'
+test_expect_success 'dump file was created with RESTORE link' '
+	tar tvf dump/RESTORE
+'
+
 test_done
diff --git a/t/t2810-kvs-garbage-collect.t b/t/t2810-kvs-garbage-collect.t
new file mode 100755
index 000000000000..06516726fd03
--- /dev/null
+++ b/t/t2810-kvs-garbage-collect.t
@@ -0,0 +1,92 @@
+#!/bin/sh
+
+test_description='Test offline KVS garbage collection'
+
+. $(dirname $0)/sharness.sh
+
+test_expect_success 'create test script' '
+	cat >runjobs.sh <<-EOT &&
+	#!/bin/bash -e
+	trap "" SIGHUP
+	flux mini submit --cc=1-10 /bin/true >/dev/null
+	flux queue drain
+	backingmod=\$(flux getattr content.backing-module)
+	flux module stats --type int --parse object_count \$backingmod
+	EOT
+	chmod +x runjobs.sh
+'
+test_expect_success 'run instance that leaves an auto dump' '
+	mkdir -p state &&
+	flux start -o,-Sstatedir=state \
+	    -o,-Scontent.dump=auto \
+	    -o,-Slog-filename=dmesg.log \
+	    ./runjobs.sh >object_count
+'
+test_expect_success 'broker logs report dump activity' '
+	grep "dumping content to" dmesg.log
+'
+test_expect_success 'dump exists and RESTORE symlink is valid' '
+	test -h state/dump/RESTORE &&
+	readlink -f state/dump/RESTORE >archive &&
+	test -f $(cat archive)
+'
+test_expect_success 'restart instance with auto restore' '
+	flux start -o,-Sstatedir=state \
+	    -o,-Scontent.restore=auto \
+	    -o,-Slog-filename=dmesg2.log \
+	    flux module stats \
+	        --type int --parse object_count content-sqlite >object_count2
+'
+test_expect_success 'broker logs report restore activity' '
+	grep "restoring content from" dmesg2.log
+'
+test_expect_success 'number of stored objects was reduced by GC' '
+	before=$(cat object_count) &&
+	after=$(cat object_count2) &&
+	test $before -gt $after
+'
+test_expect_success 'RESTORE symlink is gone' '
+	test_must_fail test -h state/dump/RESTORE
+'
+test_expect_success 'archive file remains' '
+	test -f $(cat archive)
+'
+
+#
+# Now repeat the above test with
+# - content-files backend
+# - no statedir
+# - explicitly named dump file (not auto)
+#
+test_expect_success 'run instance that leaves a named dump' '
+	flux start -o,-Slog-filename=dmesg3.log \
+	    -o,-Scontent.dump=foo.tgz \
+	    -o,-Scontent.backing-module=content-files \
+	    ./runjobs.sh >object_count3
+'
+test_expect_success 'broker logs report dump activity' '
+	grep "dumping content to" dmesg3.log
+'
+test_expect_success 'dump exists in current directory' '
+	test -f foo.tgz
+'
+test_expect_success 'no RESTORE link was created because path is explicit' '
+	test_must_fail test -h dump/RESTORE
+'
+test_expect_success 'restart instance and restore' '
+	flux start -o,-Slog-filename=dmesg4.log \
+	    -o,-Scontent.restore=foo.tgz \
+	    -o,-Scontent.backing-module=content-files \
+	    flux module stats \
+	        --type int --parse object_count content-files >object_count4
+'
+test_expect_success 'broker logs report restore activity' '
+	grep "restoring content from" dmesg4.log
+'
+test_expect_success 'number of stored objects was reduced by GC' '
+	before=$(cat object_count3) &&
+	after=$(cat object_count4) &&
+	test $before -gt $after
+'
+
+test_done