gfs2: time journal recovery steps accurately
authorAbhi Das <adas@redhat.com>
Thu, 29 Mar 2018 17:41:27 +0000 (10:41 -0700)
committerBob Peterson <rpeterso@redhat.com>
Thu, 29 Mar 2018 17:41:27 +0000 (10:41 -0700)
This patch spits out the time taken by the various steps in the
journal recover process. Previously, the journal recovery time
didn't account for finding the journal head in the log which takes
up a significant portion of time.

Signed-off-by: Abhi Das <adas@redhat.com>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
fs/gfs2/recovery.c

index b6b2589..d8b622c 100644 (file)
@@ -15,6 +15,7 @@
 #include <linux/gfs2_ondisk.h>
 #include <linux/crc32.h>
 #include <linux/crc32c.h>
+#include <linux/ktime.h>
 
 #include "gfs2.h"
 #include "incore.h"
@@ -409,12 +410,13 @@ void gfs2_recover_func(struct work_struct *work)
        struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
        struct gfs2_log_header_host head;
        struct gfs2_holder j_gh, ji_gh, thaw_gh;
-       unsigned long t;
+       ktime_t t_start, t_jlck, t_jhd, t_tlck, t_rep;
        int ro = 0;
        unsigned int pass;
        int error;
        int jlocked = 0;
 
+       t_start = ktime_get();
        if (sdp->sd_args.ar_spectator ||
            (jd->jd_jid != sdp->sd_lockstruct.ls_jid)) {
                fs_info(sdp, "jid=%u: Trying to acquire journal lock...\n",
@@ -446,6 +448,7 @@ void gfs2_recover_func(struct work_struct *work)
                fs_info(sdp, "jid=%u, already locked for use\n", jd->jd_jid);
        }
 
+       t_jlck = ktime_get();
        fs_info(sdp, "jid=%u: Looking at journal...\n", jd->jd_jid);
 
        error = gfs2_jdesc_check(jd);
@@ -455,13 +458,12 @@ void gfs2_recover_func(struct work_struct *work)
        error = gfs2_find_jhead(jd, &head);
        if (error)
                goto fail_gunlock_ji;
+       t_jhd = ktime_get();
 
        if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
                fs_info(sdp, "jid=%u: Acquiring the transaction lock...\n",
                        jd->jd_jid);
 
-               t = jiffies;
-
                /* Acquire a shared hold on the freeze lock */
 
                error = gfs2_glock_nq_init(sdp->sd_freeze_gl, LM_ST_SHARED,
@@ -495,6 +497,7 @@ void gfs2_recover_func(struct work_struct *work)
                        goto fail_gunlock_thaw;
                }
 
+               t_tlck = ktime_get();
                fs_info(sdp, "jid=%u: Replaying journal...\n", jd->jd_jid);
 
                for (pass = 0; pass < 2; pass++) {
@@ -509,9 +512,14 @@ void gfs2_recover_func(struct work_struct *work)
                clean_journal(jd, &head);
 
                gfs2_glock_dq_uninit(&thaw_gh);
-               t = DIV_ROUND_UP(jiffies - t, HZ);
-               fs_info(sdp, "jid=%u: Journal replayed in %lus\n",
-                       jd->jd_jid, t);
+               t_rep = ktime_get();
+               fs_info(sdp, "jid=%u: Journal replayed in %lldms [jlck:%lldms, "
+                       "jhead:%lldms, tlck:%lldms, replay:%lldms]\n",
+                       jd->jd_jid, ktime_ms_delta(t_rep, t_start),
+                       ktime_ms_delta(t_jlck, t_start),
+                       ktime_ms_delta(t_jhd, t_jlck),
+                       ktime_ms_delta(t_tlck, t_jhd),
+                       ktime_ms_delta(t_rep, t_tlck));
        }
 
        gfs2_recovery_done(sdp, jd->jd_jid, LM_RD_SUCCESS);