mxqd: Fix reaper - pid might not be known when recovering so reap by …

…job_id

mxqd.c

@@ -1787,7 +1787,7 @@ static int fspool_process_file(struct mxq_server *server,char *filename, uint64_
 
     mx_log_info("job finished (via fspool) : job %lu pid %d status %d", job_id, pid, status);
 
-    jlist = server_remove_job_list_by_pid(server, pid);
+    jlist = server_remove_job_list_by_job_id(server, job_id);
     if (!jlist) {
         mx_log_warning("fspool_process_file: %s : job unknown on server", filename);
         return -(errno=ENOENT);
@@ -1921,7 +1921,7 @@ static int lost_scan_one(struct mxq_server *server)
                     mx_log_warning("pid %u: process is gone. cancel job %lu",
                                 jlist->job.host_pid,
                                 jlist->job.job_id);
-                    server_remove_job_list_by_pid(server, job->host_pid);
+                    server_remove_job_list_by_job_id(server, job->job_id);
 
                     job->job_status = MXQ_JOB_STATUS_UNKNOWN;
 

mxqd_control.c

@@ -227,6 +227,19 @@ struct mxq_job_list *server_remove_job_list_by_pid(struct mxq_server *server, pi
     return jlist;
 }
 
+struct mxq_job_list *server_remove_job_list_by_job_id(struct mxq_server *server, uint64_t job_id)
+{
+    struct mxq_job_list *jlist;
+
+    assert(server);
+
+    jlist = server_get_job_list_by_job_id(server, job_id);
+    if (jlist) {
+        job_list_remove_self(jlist);
+    }
+    return jlist;
+}
+
 static struct mxq_user_list *_user_list_find_by_uid(struct mxq_user_list *ulist, uint32_t uid)
 {
     for (; ulist; ulist = ulist->next) {

mxqd_control.h

@@ -6,6 +6,8 @@
 #include "mxq_job.h"
 
 struct mxq_job_list   *server_remove_job_list_by_pid(struct mxq_server *server, pid_t pid);
+struct mxq_job_list   *server_remove_job_list_by_job_id(struct mxq_server *server, uint64_t job_id);
+
 struct mxq_group_list *server_update_group(struct mxq_server *server, struct mxq_group *group);
 
 struct mxq_group_list *server_get_group_list_by_group_id(struct mxq_server *server, uint64_t group_id);