scoutfs-tests: add createmany-rename-large-dir

Add a test that randomly renames entries in a single large directory.
This has caught bugs in the reservation of allocator resources for
client transactions.

Signed-off-by: Zach Brown <zab@versity.com>

tests/golden/createmany-rename-large-dir

@@ -0,0 +1,2 @@
+== create large directory with 1220608 files
+== randomly renaming 5000 files

tests/sequence

@@ -13,6 +13,7 @@ lock-pr-cw-conflict.sh
 lock-revoke-getcwd.sh
 createmany-parallel.sh
 createmany-large-names.sh
+createmany-rename-large-dir.sh
 stage-release-race-alloc.sh
 basic-posix-consistency.sh
 dirent-consistency.sh

tests/tests/createmany-rename-large-dir.sh

@@ -0,0 +1,39 @@
+#
+# We've had bugs where client transactions weren't properly being
+# committed as their allocators ran out of resources for the amount of
+# log btree dirtying they'd need to do for the size of their dirty item
+# cache pages.
+#
+# This stresses those heuristics by trying to maximize the number of
+# btree blocks dirtied for each dirty item cache page.  We create an
+# enormous directory and then randomly rename entries in it.
+#
+# With the bad client commit heuristics this would reliably fail before
+# a few thousand renames.
+#
+
+t_require_commands createmany mv mkdir
+
+# 298 dirents per 64k block, 4096 max meta avail -- often fewer
+NR=$((298 * 4096))
+RENAMES=5000
+
+DIR="$T_D0/dir/"
+
+echo "== create large directory with $NR files"
+t_quiet mkdir -p "$DIR"
+./src/createmany -o "$DIR/f-" $NR > "$T_TMP.createmany.stdout"
+
+echo "== randomly renaming $RENAMES files"
+for i in $(seq 1 $RENAMES); do
+        rnd=$(((1$RANDOM$RANDOM) % NR))
+        orig="$DIR/f-$rnd"
+        tmp="$DIR/f-$rnd.$i"
+
+        mv "$orig" "$tmp"
+        mv "$tmp" "$orig"
+done
+
+t_quiet rm -rf "$DIR"
+
+t_pass