Hello,
Attached, you will find a patch that I am proposing to release. Feedback
would be appreciated.
Best regards,
Kern
On Sunday 10 February 2008 04.52:26 Peter Much wrote:
> Abstract:
> ---------
> Migration jobs do not insert the jobmedia.startfile attribute
> in the catalog.
>
>
> Impact:
> -------
> Very slow restore.
>
>
> Example:
> --------
> bacula=> select jobid,name,type,jobfiles,poolid,priorjobid from job
> where jobid=1728 or jobid=1729 or jobid=1748 or jobid=1749
> order by jobid;
> jobid | name | type | jobfiles | poolid | priorjobid
> -------+--------------------+------+----------+--------+------------
> 1728 | HomeDirsDisp | B | 4398 | 5 | 0
> 1729 | HomeDirsDisp | M | 4398 | 2 | 0
> 1748 | MoveToTapeInternal | g | 0 | 5 | 0
> 1749 | HomeDirsDisp | B | 4398 | 5 | 1729
> (4 rows)
>
> -> job 1728 saves directly to pool 5. Job 1729 saves the same data
> to pool 2, and migrating job 1748/1749 moves it on to pool 5.
>
> bacula=> select jobid,firstindex,lastindex,startfile,endfile,
> startblock,endblock from jobmedia
> where jobid=1728 or jobid=1729 or jobid=1748 or jobid=1749
> order by jobid;
> jobid | firstindex | lastindex | startfile | endfile | startblock |
> endblock
> -------+------------+-----------+-----------+---------+------------+-------
>---- 1728 | 1 | 4398 | 55 | 55 | 0 |
> 3466 1729 | 1 | 4398 | 0 | 0 | 217 |
> 223662044 1749 | 1 | 4398 | 0 | 56 | 0
> | 3466 (3 rows)
>
> -> Job 1728 sets startfile attribute, migrate job 1749 sets just 0.
>
>
> Comment:
> --------
> 1. I have tried to fix the StartFile issue, so that my restores
> run better. I have not further checked how the other data are
> derived on Migration and if there may be more problems.
> Usually I do such, but I was tired at that point. ;)
>
> 2. The existence of jcr->dcr is checked twice in that function,
> once right after entry, and once at label "ok_out:". Could it
> be something "stealing" this object inflight?
> If so, then my earlier using of "dev" as a shorthand for
> jcr->dcr->dev is illegitim, and things need to be written
> more cumbersomely.
> (I have obviousely not gone into the layout of the whole
> software conception already; therefore these patches should
> be considered more as suggestions to where the problem seems
> to arise, and not so much as readymade fixes!)
>
> Patch:
> ------
> --- src/stored/mac.c.orig Wed Oct 3 13:36:47 2007
> +++ src/stored/mac.c Sun Feb 10 04:28:54 2008
> @@ -89,7 +89,7 @@
> }
>
> Dmsg3(200, "Found %d volumes names for %s. First=%s\n",
> jcr->NumReadVolumes, - jcr->VolList->VolumeName, Type);
> + Type, jcr->VolList->VolumeName);
>
> /* Ready devices for reading and writing */
> if (!acquire_device_for_read(jcr->read_dcr) ||
> @@ -98,8 +98,26 @@
> goto bail_out;
> }
>
> - Dmsg2(200, "===== After acquire pos %u:%u\n", jcr->dcr->dev->file,
> jcr->dcr->dev->block_num); + dev = jcr->dcr->dev;
> + Dmsg2(200, "===== After acquire pos %u:%u\n", dev->file,
> dev->block_num);
>
> + /* On MIGRATE we must setup some dcr jobmedia data. On BACKUP
> + * this is done in write_session_label().
> + */
> + switch(jcr->JobType) {
> + case JT_MIGRATE:
> + if (dev->is_tape()) {
> + jcr->dcr->StartBlock = dev->block_num;
> + jcr->dcr->StartFile = dev->file;
> + } else {
> + jcr->dcr->StartBlock = (uint32_t)dev->file_addr;
> + jcr->dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
> + }
> + break;
> + default:
> + break;
> + }
> +
> set_jcr_job_status(jcr, JS_Running);
> dir_send_job_status(jcr);
>
> @@ -117,7 +135,6 @@
>
> ok_out:
> if (jcr->dcr) {
> - dev = jcr->dcr->dev;
> if (ok || dev->can_write()) {
> /* Flush out final partial block of this session */
> if (!write_block_to_device(jcr->dcr)) {
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> Bacula-devel mailing list
> Bacula-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/bacula-devel
This patch fixes a migration bug that always has a zero index entry
(JobMedia record) as the first entry. This causes Bacula to search
for the first record during a restore rather than seek directly to
it.
Apply this patch to Bacula 2.2.8 (and possibly any prior 2.2.x version) with:
cd <bacula-source>
patch -p0 <2.2.8-jobmedia.patch
./configure <your-options>
make
...
make install
Index: src/stored/device.c
===================================================================
--- src/stored/device.c (revision 6391)
+++ src/stored/device.c (working copy)
@@ -1,7 +1,7 @@
/*
Bacula® - The Network Backup Solution
- Copyright (C) 2000-2008 Free Software Foundation Europe e.V.
+ Copyright (C) 2000-2007 Free Software Foundation Europe e.V.
The main author of Bacula is Kern Sibbald, with contributions from
many others, a complete list can be found in the file AUTHORS.
@@ -200,6 +200,19 @@
return ok; /* device locked */
}
+void set_start_vol_position(DCR *dcr)
+{
+ DEVICE *dev = dcr->dev;
+ /* Set new start position */
+ if (dev->is_tape()) {
+ dcr->StartBlock = dev->block_num;
+ dcr->StartFile = dev->file;
+ } else {
+ dcr->StartBlock = (uint32_t)dev->file_addr;
+ dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
+ }
+}
+
/*
* We have a new Volume mounted, so reset the Volume parameters
* concerning this job. The global changes were made earlier
@@ -208,24 +221,11 @@
void set_new_volume_parameters(DCR *dcr)
{
JCR *jcr = dcr->jcr;
- DEVICE *dev = dcr->dev;
if (dcr->NewVol && !dir_get_volume_info(dcr, GET_VOL_INFO_FOR_WRITE)) {
Jmsg1(jcr, M_ERROR, 0, "%s", jcr->errmsg);
}
- /* Set new start/end positions */
- if (dev->is_tape()) {
- dcr->StartBlock = dev->block_num;
- dcr->StartFile = dev->file;
- } else {
- dcr->StartBlock = (uint32_t)dev->file_addr;
- dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
- }
- /* Reset indicies */
- dcr->VolFirstIndex = 0;
- dcr->VolLastIndex = 0;
+ set_new_file_parameters(dcr);
jcr->NumWriteVolumes++;
- dcr->NewVol = false;
- dcr->WroteVol = false;
}
/*
@@ -235,16 +235,8 @@
*/
void set_new_file_parameters(DCR *dcr)
{
- DEVICE *dev = dcr->dev;
+ set_start_vol_position(dcr);
- /* Set new start/end positions */
- if (dev->is_tape()) {
- dcr->StartBlock = dev->block_num;
- dcr->StartFile = dev->file;
- } else {
- dcr->StartBlock = (uint32_t)dev->file_addr;
- dcr->StartFile = (uint32_t)(dev->file_addr >> 32);
- }
/* Reset indicies */
dcr->VolFirstIndex = 0;
dcr->VolLastIndex = 0;
Index: src/stored/mac.c
===================================================================
--- src/stored/mac.c (revision 6391)
+++ src/stored/mac.c (working copy)
@@ -1,15 +1,7 @@
/*
- * SD -- mac.c -- responsible for doing
- * migration, archive, and copy jobs.
- *
- * Kern Sibbald, January MMVI
- *
- * Version $Id$
- */
-/*
Bacula® - The Network Backup Solution
- Copyright (C) 2006-2006 Free Software Foundation Europe e.V.
+ Copyright (C) 2006-2008 Free Software Foundation Europe e.V.
The main author of Bacula is Kern Sibbald, with contributions from
many others, a complete list can be found in the file AUTHORS.
@@ -33,6 +25,14 @@
(FSFE), Fiduciary Program, Sumatrastrasse 25, 8006 Zürich,
Switzerland, email:[EMAIL PROTECTED]
*/
+/*
+ * SD -- mac.c -- responsible for doing
+ * migration, archive, and copy jobs.
+ *
+ * Kern Sibbald, January MMVI
+ *
+ * Version $Id$
+ */
#include "bacula.h"
#include "stored.h"
@@ -108,6 +108,7 @@
jcr->dcr->VolFirstIndex = jcr->dcr->VolLastIndex = 0;
jcr->run_time = time(NULL);
+ set_start_vol_position(jcr->dcr);
ok = read_records(jcr->read_dcr, record_cb, mount_next_read_volume);
goto ok_out;
Index: src/stored/protos.h
===================================================================
--- src/stored/protos.h (revision 6391)
+++ src/stored/protos.h (working copy)
@@ -126,6 +126,7 @@
bool open_device(DCR *dcr);
bool first_open_device(DCR *dcr);
bool fixup_device_block_write_error(DCR *dcr);
+void set_start_vol_position(DCR *dcr);
void set_new_volume_parameters(DCR *dcr);
void set_new_file_parameters(DCR *dcr);
bool is_device_unmounted(DEVICE *dev);
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
Bacula-devel mailing list
Bacula-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/bacula-devel
