ci: Move stale job cleanup to get-test-job.c

It's more efficient to do it there, and now we emit a log message when
we delete a stale test job.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

2 files changed, 41 insertions, 27 deletions

diff --git a/ci/get-test-job.sh b/ci/get-test-job.sh
index bf64c6b..8347b08 100755
--- a/ci/get-test-job.sh
+++ b/ci/get-test-job.sh
@@ -2,24 +2,9 @@
 
 [[ -f ~/.ktestrc ]] && . ~/.ktestrc
 
-# Clean stale test jobs:
-
-cd $JOBSERVER_OUTPUT_DIR
-
-if [[ ! -f stale-job-cleanup ]]; then
-    touch stale-job-cleanup
-fi
-
-if [[ $(find stale-job-cleanup -mmin +5) ]]; then
-    echo -n "Cleaning stale jobs.. " >&2
-    find -size 0 -cmin +180 |xargs rm -f > /dev/null
-    touch stale-job-cleanup
-    echo " done" >&2
-fi
-
 cd $JOBSERVER_HOME/linux
 flock --nonblock .git_fetch.lock git fetch --all > /dev/null
 
 make -C ~/ktest/lib get-test-job 1>&2
 
-~/ktest/lib/get-test-job -b ~/BRANCHES-TO-TEST -o $JOBSERVER_OUTPUT_DIR
+~/ktest/lib/get-test-job -b ~/BRANCHES-TO-TEST -o $JOBSERVER_OUTPUT_DIR/c
diff --git a/lib/get-test-job.c b/lib/get-test-job.c
index f45b947..aab6a2f 100644
--- a/lib/get-test-job.c
+++ b/lib/get-test-job.c
@@ -10,6 +10,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <sys/stat.h>
+#include <sys/time.h>
 #include <sys/types.h>
 #include <unistd.h>
 
@@ -158,6 +159,31 @@ static char *slashes_to_dots(const char *str)
 	return ret;
 }
 
+static bool __lockfile_exists(const char *commitdir,
+			      const char *testdir,
+			      const char *lockfile,
+			      bool create)
+{
+	if (!create) {
+		return access(lockfile, F_OK) == 0;
+	} else {
+		bool exists;
+
+		if (mkdir(commitdir, 0755) < 0 && errno != EEXIST)
+			die("error creating %s", commitdir);
+
+		if (mkdir(testdir, 0755) < 0 && errno != EEXIST)
+			die("error creating %s", testdir);
+
+		int fd = open(lockfile, O_RDWR|O_CREAT|O_EXCL, 0644);
+		exists = fd < 0;
+		if (!exists)
+			close(fd);
+
+		return exists;
+	}
+}
+
 static bool lockfile_exists(const char *commit,
 			    const char *test_path,
 			    const char *subtest,
@@ -168,21 +194,24 @@ static bool lockfile_exists(const char *commit,
 	char *commitdir = mprintf("%s/%s", outdir, commit);
 	char *testdir = mprintf("%s/%s.%s", commitdir, test_name, subtest_mangled);
 	char *lockfile = mprintf("%s/status", testdir);
+	struct timeval now;
+	struct stat statbuf;
 	bool exists;
 
-	if (!create) {
-		exists = access(lockfile, F_OK) == 0;
-	} else {
-		if (mkdir(commitdir, 0755) < 0 && errno != EEXIST)
-			die("error creating %s", commitdir);
+	gettimeofday(&now, NULL);
 
-		if (mkdir(testdir, 0755) < 0 && errno != EEXIST)
-			die("error creating %s", testdir);
+	exists = __lockfile_exists(commitdir, testdir, lockfile, create);
 
-		int fd = open(lockfile, O_RDWR|O_CREAT|O_EXCL, 0644);
-		exists = fd < 0;
-		if (!exists)
-			close(fd);
+	if (exists &&
+	    !stat(lockfile, &statbuf) &&
+	    !statbuf.st_size &&
+	    S_ISREG(statbuf.st_mode) &&
+	    statbuf.st_ctime + 20 * 60 < now.tv_sec &&
+	    !unlink(lockfile)) {
+		fprintf(stderr, "Deleting stale test job %s %s %s (%lu minutes old)\n",
+			commit, test_name, subtest,
+			(now.tv_sec - statbuf.st_ctime) / 60);
+		exists = false;
 	}
 
 	free(lockfile);