HW - DUMP

Dump - Core

AIX generates a system dump when a severe error occurs. A system dump creates a picture of the system's memory contents. If the AIX kernel crashes kernel data is written to the primary dump device. After a kernel crash AIX must be rebooted. During the next boot, the dump is copied into a dump directory (default is /var/adm/ras). The dump file name is vmcore.x (x indicates a number, e.g. vmcore.0)

When installing the operating system, the dump device is automatically configured. By default, the primary device is /dev/hd6, which is a paging logical volume, and the secondary device is /dev/sysdumpnull.

A rule of thumb is when a dump is created, it is about 1/4 of the size of real memory. The command "sysdumpdev -e" will also provide an estimate of the dump space needed for your machine. (Estimation can differ at times with high load, as kernel space is higher at that time.)

When a system dump is occurring, the dump image is not written to disk in mirrored form. A dump to a mirrored lv results in an inconsistent dump and therefore, should be avoided. The logic behind this fact is that if the mirroring code itself were the cause of the system crash, then trusting the same code to handle the mirrored write would be pointless. Thus, mirroring a dump device is a waste of resources and is not recommended.

Since the default dump device is the primary paging lv, you should create a separate dump lv, if you mirror your paging lv (which is suggested.)If a valid secondary dump device exists and the primary dump device cannot be reached, the secondary dump device will accept the dump information intended for the primary dump device.

IBM recommendation:
All I can recommend you is to force a dump the next time the problem should occur. This will enable us to check which process was hanging or what caused the system to not respond any more. You can do this via the HMC using the following steps:
   Operations -> Restart -> Dump
As a general recommendation you should always force a dump if a system is hanging. There are only very few cases in which we can determine the reason for a hanging system without having a dump available for analysis.

-------------------------------------------

Traditional vs Firmware-assisted dump:

Up to POWER5 only traditioanl dumps were available, and the introduction of the POWER6 processor-based systems allowed system  dumps to be firmware assisted. When performing a firmware-assisted dump, system memory is frozen and the partition rebooted, which allows a new instance of the operating system to complete the dump.

Traditional dump: it is generated before partition is rebooted.
(When system crashed, memory content is trying to be copied at that moment to dump device)

Firmware-assisted dump: it takes place when the partition is restarting.
(When system crashed, memory is frozen, and by hypervisor (firmware) new memory space is allocated in RAM, and the contents of memory is copied there. Then during reboot it is copied from this new memory area to the dump device.)

Firmware-assisted dump offers improved reliability over the traditional dump, by rebooting the partition and using a new kernel to dump data from the previous kernel crash.

When an administrator attempts to switch from a traditional to firmware-assisted system dump, system memory is checked against the firmware-assisted system dump memory requirements. If these memory requirements are not met, then the "sysdumpdev -t" command output reports the required minimum system memory to allow for firmware-assisted dump to be configured. Changing from traditional to firmware-assisted dump requires a reboot of the partition for the dump changes to take effect.

Firmware-assisted system dumps can be one of these types:

Selective memory dump: Selective memory dumps are triggered by or use of AIX instances that must be dumped.
Full memory dump: The whole partition memory is dumped without any interaction with an AIX instance that is failing.

-------------------------------------------


Use the sysdumpdev command to query or change the primary or secondary dump devices.
    - Primary:    usually used when you wish to save the dump data
    - Secondary: can be used to discard dump data (that is, /dev/sysdumpnull)


Flags for sysdumpdev command:
    -l                list the current dump destination
    -e                estimates the size of the dump (in bytes)
    -p                primary
    -s                secondary
    -P                make change permanent
    -C                turns on compression
    -c                turns off compression
    -L                shows info about last dump
    -K                turns on: alway allow system dump

sysdumpdev -P -p /dev/dumpdev    change the primary dumpdevice permanently to /dev/dumpdev

root@aix1: /root # sysdumpdev -l
primary              /dev/dumplv
secondary            /dev/sysdumpnull
copy directory       /var/adm/ras
forced copy flag     TRUE
always allow dump    TRUE          <--if it is on FALSE then in smitty sysdumpdev it can be change
dump compression     ON            <--if it is on OFF then sysdumpdev -C changes it to ON-ra (-c changes it to OFF)
  


Other commands:

sysdumpstart            starts a dump (smitty dump)(it will do a reboot as well)
kdb                     it analysis the dump
/usr/lib/ras/dumpcheck  checks if dump device and copy directory are able to receive the system dump
If dump device is a paging space, it verifies if enough free space exists in the copy dir to hold the dump
If dump device is a logical volume, it verifies it is large enough to hold a dump
(man dumpcheck)

-------------------------------------------
 SNAP:

snap
    -a                copies all system config. information to /tmp/ibmsupt directory tree
    -c                creates a  compressed tar image (snap.tar.Z) of all files in the /tmp/ibmsupt
    -g                gather general information

    -e                for HACMP, it runs clverification and gathers the data creating a snap

1. snap -r            <--removes old snap from /tmp/ibmsupt
2. snap -gc           <--creates a new snap file

Reading a compressed snap file:
1. snap -ac                     <--creates a compressed snap file (/tmp/ibmsupt/snap.pax.Z)
2. uncompress snap.pax.Z        <--uncompresses it, we will have a snap.pax file
3. pax -rvf snap.pax            <--unpack files, after files can be read


-------------------------------------------

Creating a dump device

1. sysdumpdev -e                                    <--shows an estimation, how much space is required for a dump
2. mklv -t sysdump -y lg_dumplv rootvg 3  hdisk0    <--it creates a sysdump lv with 3 PPs
3. sysdumpdev -Pp /dev/lg_dumplv                    <--making it as a primary device (system will use this lv now for dumps)

-------------------------------------------

System dump initiaded by a user

!!!reboot will take place automatically!!!

1. sysdumpstart -p              <--initiates a dump to the primary device
(Reboot will be done automatically)
(If a dedicated dump device is used, user initiated dumps are not copied automatically to copy directory.)
(If paging space is used for dump, then dump will be copied automatically to /var/adm/ras)
2. sysdumpdev -L               <--shows dump took place on the primary device, time, size ... (errpt will show as well)
3. savecore -d /var/adm/ras    <--copy last dump from system dump device to directory /var/adm/ras (if paging space is used this is not needed)

-------------------------------------------

How to move dumplv to another disk:

We want to move from hdisk1 to hdisk0:

1. lslv -l dumplv                      <--checking which disk
2. sysdumpdev -l                       <--checking sysdump device (primary was here /dev/dumplv)
3. sysdumpdev -Pp /dev/sysdumpnull     <--changing primary to sysdumpnull (secondary, it is a null device) (lsvg -l roovg shows closed state)
4. migratepv -l dumplv hdisk1 hdisk0   <--moving it from hdisk1 to hdisk0
5. sysdumpdev -Pp /dev/dumplv          <--changing back to the primary device


-------------------------------------------
The largest dump device is too small: (LABEL: DMPCHK_TOOSMALL IDENT)

1. Dumpcheck runs from crontab
# crontab -l | grep dump
0 15 * * * /usr/lib/ras/dumpcheck >/dev/null 2>&1

2. Check if there are any errors:
# errpt
IDENTIFIER TIMESTAMP T C RESOURCE_NAME DESCRIPTION
E87EF1BE 0703150008 P O dumpcheck The largest dump device is too small.
E87EF1BE 0702150008 P O dumpcheck The largest dump device is too small.

3. If you find new error message, find dumplv:
# sysdumpdev -l
primary /dev/dumplv
secondary /dev/sysdumpnull
copy directory /var/adm/ras
forced copy flag TRUE
always allow dump TRUE
dump compression ON
List dumplv form rootvg:

# lsvg -l rootvg|grep dumplv
dumplv dump 8 8 1 open/syncd N/A

4. Extend with 1 PP
# extendlv dumplv 1
dumplv dump 9 9 1 open/syncd N/A

Run problem check at the end
OK -> done
Not OK -> Extend with 1 PP again.
-------------------------------------------
changing the autorestart attribute of the systemdump:
(smitty chgsys as well)

1.lsattr -El sys0 -a autorestart
autorestart true Automatically REBOOT system after a crash True

2.chdev -l sys0 -a autorestart=false
sys0 changed

----------------------------------------------

CORE FILE:

errpt shows which program, if not:
- use the “strings”  command (for example: ”strings core | grep _=”)
- or the lquerypv command: (for example: “lquerypv -h core 6b0 64”)

man syscorepath
syscorepath -p /tmp
syscorepath -g