[RHEL] How to use a new partition in RHEL6 without reboot?

URL복사							공유하기
|
기존에 RHEL5을 비롯한 이전버전의 Linux에서 변경한 파티션정보를 커널이 다시 읽도록 하여 파티션테이블을 갱신하는데 사용하던 partprobe 명령의 기능을 RHEL6에서는 더 이상 사용할 수 없습니다. 버그가 아니라 데이터의 안정성을 위해서 사용중인(eg,mount) 장치에서는 갱신된 파티션을 커널에 적용되지 않도록 parted 패키지가 변경이 되었다고 합니다. (partprobe는 parted패키지에서 제공하는 유틸) 이유는 RHEL5를 비롯한 이전버전의 parted 패키지에서는 변경된 파티션 정보를 커널에 적용하는 과정에서 오류를 체크하는 부분이 없었기 때문입니다. partprobe가 종종 동작되지 않을 때도, 그 발생한 오류는 사용자에게 보고되지 않기 때문에 사용자는 파티션 작업이 정상적으로 적용되었다고 착각할 수 있지만, 실제 장치에서는 파티션 변경 내용이 적용되지 않아 데이터 입출력 간의 중첩쓰기로 인한 데이터 손실 등이 우려되므로, RHEL6에서는 데이터 안정성을 위해 사용중인 장치의 경우 변경된 파티션 테이블을 갱신하지 못하도록 수정되었습니다. 따라서 partprobe를 이용하여 변경된 파티션 정보를 다시 읽어 커널의 파티션테이블을 갱신하도록 하려면 mount 되어 있는 장치를 모두 umount 한 후 partprobe를 사용하거나 서버를 리부팅하여야 합니다. 하지만 partprobe를 사용하지 않고도 커널의 파티션테이블을 갱신할 수 있는 방법이 있는데, partx 명령의 기능을 사용하면 사용중인 장치를 umount하거나 리부팅하지 않고도 커널의 파티션테이블을 갱신할 수 있습니다. (partx는 util-linux-ng 패키지에서 제공하는 유틸) # partx -v -a /dev/sda Issue Added a new partition to a disk that have some existing partitions already mounted. Can the new partition be used without reboot in RHEL6? Environment Red Hat Enterprise Linux (RHEL) 6 Resolution partprobe was commonly used in RHEL 5 to inform the OS of partition table changes on the disk. In RHEL 6, it will only trigger the OS to update the partitions on a disk that none of its partitions are in use (e.g. mounted). If any partition on a disk is in use, partprobe will not trigger the OS to update partitions in the system because it is considered unsafe in some situations. So in general we would suggest: Unmount all the partitions of the disk before modifying the partition table on the disk, and then runpartprobe to update the partitions in system. If this is not possible (e.g. the mounted partition is a system partition), reboot the system after modifying the partition table. The partitions information will be re-read after reboot. If a new partition was added and none of the existing partitions were modified, consider using the partxcommand to update the system partition table. Do note that the partx command does not do much checking between the new and the existing partition table in the system and assumes the user knows what they are are doing. So it can corrupt the data on disk if the existing partitions are modified or the partition table is not set correctly. So use at one's own risk. For example, a partition #1 is an existing partition and a new partition #2 is already added in /dev/sdb by fdisk. Here we use partx -v -a /dev/sdb to add the new partition to the system: # ls /dev/sdb* /dev/sdb /dev/sdb1 List the partition table of disk: # partx -l /dev/sdb # 1: 63- 505007 ( 504945 sectors, 258 MB) # 2: 505008- 1010015 ( 505008 sectors, 258 MB) # 4: 0- -1 ( 0 sectors, 0 MB) # 3: 0- -1 ( 0 sectors, 0 MB) Read disk and try to add all partitions to the system: # partx -v -a /dev/sdb device /dev/sdb: start 0 size 2097152 gpt: 0 slices ors, 258 MB) dos: 4 slices # 1: 63- 505007 ( 504945 sec t # 2: 505008- 1010015 ( 505008 sectors, 258 MB) 0- -1 ( 0 sectors, 0 MB) BLK # 3: 0- -1 ( 0 sectors, 0 MB) # 4 :PG: Device or resource busy error adding partition 1 (These last 2 lines are normal in this case because partition 1 is already added in the system before partition 2 is added) Check that we have device nodes for /dev/sdb itself and the partitions on it: # ls /dev/sdb* /dev/sdb /dev/sdb1 /dev/sdb2 Comments Ryan Shannon August 8 2013 at 2:22 PMPermalink The question is how to use a new partition without rebooting, and the solution is to take the disk out of service (equivalent to rebooting, from a downtime perspective) or reboot? Dynamic resizing of disks is very common in SAN and VMWare environments, and this "enhancement" to RHEL 6 that disables dynamic updates to active partition tables is ill advised, in my opinion. gleb.shulman August 16 2013 at 7:29 PMPermalink The solution contains non-reboot way too using partx. Just experimented with partitions merge: - from /dev/sda - removed partition 9 - removed partition 8 - recreated partition 8 twice big - issued partx -d -nr 8-9 /dev/sda - issued partx -a /dev/sda - checked /proc/partitions - kernel now is seeing /dev/sda8 only and reports new size.   If it will work without surprises, then would think that was an enhancement actually. Partprobe became honestly advisable and 100% safe and split out its new risky "partx" twin-brother. ^IMO

URL복사							공유하기
|
기존에 RHEL5을 비롯한 이전버전의 Linux에서 변경한 파티션정보를 커널이 다시 읽도록 하여 파티션테이블을 갱신하는데 사용하던 partprobe 명령의 기능을 RHEL6에서는 더 이상 사용할 수 없습니다. 버그가 아니라 데이터의 안정성을 위해서 사용중인(eg,mount) 장치에서는 갱신된 파티션을 커널에 적용되지 않도록 parted 패키지가 변경이 되었다고 합니다. (partprobe는 parted패키지에서 제공하는 유틸) 이유는 RHEL5를 비롯한 이전버전의 parted 패키지에서는 변경된 파티션 정보를 커널에 적용하는 과정에서 오류를 체크하는 부분이 없었기 때문입니다. partprobe가 종종 동작되지 않을 때도, 그 발생한 오류는 사용자에게 보고되지 않기 때문에 사용자는 파티션 작업이 정상적으로 적용되었다고 착각할 수 있지만, 실제 장치에서는 파티션 변경 내용이 적용되지 않아 데이터 입출력 간의 중첩쓰기로 인한 데이터 손실 등이 우려되므로, RHEL6에서는 데이터 안정성을 위해 사용중인 장치의 경우 변경된 파티션 테이블을 갱신하지 못하도록 수정되었습니다. 따라서 partprobe를 이용하여 변경된 파티션 정보를 다시 읽어 커널의 파티션테이블을 갱신하도록 하려면 mount 되어 있는 장치를 모두 umount 한 후 partprobe를 사용하거나 서버를 리부팅하여야 합니다. 하지만 partprobe를 사용하지 않고도 커널의 파티션테이블을 갱신할 수 있는 방법이 있는데, partx 명령의 기능을 사용하면 사용중인 장치를 umount하거나 리부팅하지 않고도 커널의 파티션테이블을 갱신할 수 있습니다. (partx는 util-linux-ng 패키지에서 제공하는 유틸) # partx -v -a /dev/sda Issue Added a new partition to a disk that have some existing partitions already mounted. Can the new partition be used without reboot in RHEL6? Environment Red Hat Enterprise Linux (RHEL) 6 Resolution partprobe was commonly used in RHEL 5 to inform the OS of partition table changes on the disk. In RHEL 6, it will only trigger the OS to update the partitions on a disk that none of its partitions are in use (e.g. mounted). If any partition on a disk is in use, partprobe will not trigger the OS to update partitions in the system because it is considered unsafe in some situations. So in general we would suggest: Unmount all the partitions of the disk before modifying the partition table on the disk, and then runpartprobe to update the partitions in system. If this is not possible (e.g. the mounted partition is a system partition), reboot the system after modifying the partition table. The partitions information will be re-read after reboot. If a new partition was added and none of the existing partitions were modified, consider using the partxcommand to update the system partition table. Do note that the partx command does not do much checking between the new and the existing partition table in the system and assumes the user knows what they are are doing. So it can corrupt the data on disk if the existing partitions are modified or the partition table is not set correctly. So use at one's own risk. For example, a partition #1 is an existing partition and a new partition #2 is already added in /dev/sdb by fdisk. Here we use partx -v -a /dev/sdb to add the new partition to the system: # ls /dev/sdb* /dev/sdb /dev/sdb1 List the partition table of disk: # partx -l /dev/sdb # 1: 63- 505007 ( 504945 sectors, 258 MB) # 2: 505008- 1010015 ( 505008 sectors, 258 MB) # 4: 0- -1 ( 0 sectors, 0 MB) # 3: 0- -1 ( 0 sectors, 0 MB) Read disk and try to add all partitions to the system: # partx -v -a /dev/sdb device /dev/sdb: start 0 size 2097152 gpt: 0 slices ors, 258 MB) dos: 4 slices # 1: 63- 505007 ( 504945 sec t # 2: 505008- 1010015 ( 505008 sectors, 258 MB) 0- -1 ( 0 sectors, 0 MB) BLK # 3: 0- -1 ( 0 sectors, 0 MB) # 4 :PG: Device or resource busy error adding partition 1 (These last 2 lines are normal in this case because partition 1 is already added in the system before partition 2 is added) Check that we have device nodes for /dev/sdb itself and the partitions on it: # ls /dev/sdb* /dev/sdb /dev/sdb1 /dev/sdb2 Comments Ryan Shannon August 8 2013 at 2:22 PMPermalink The question is how to use a new partition without rebooting, and the solution is to take the disk out of service (equivalent to rebooting, from a downtime perspective) or reboot? Dynamic resizing of disks is very common in SAN and VMWare environments, and this "enhancement" to RHEL 6 that disables dynamic updates to active partition tables is ill advised, in my opinion. gleb.shulman August 16 2013 at 7:29 PMPermalink The solution contains non-reboot way too using partx. Just experimented with partitions merge: - from /dev/sda - removed partition 9 - removed partition 8 - recreated partition 8 twice big - issued partx -d -nr 8-9 /dev/sda - issued partx -a /dev/sda - checked /proc/partitions - kernel now is seeing /dev/sda8 only and reports new size.   If it will work without surprises, then would think that was an enhancement actually. Partprobe became honestly advisable and 100% safe and split out its new risky "partx" twin-brother. ^IMO

기존에 RHEL5을 비롯한 이전버전의 Linux에서 변경한 파티션정보를 커널이 다시 읽도록 하여 파티션테이블을 갱신하는데 사용하던 partprobe 명령의 기능을 RHEL6에서는 더 이상 사용할 수 없습니다.

버그가 아니라 데이터의 안정성을 위해서 사용중인(eg,mount) 장치에서는 갱신된 파티션을 커널에 적용되지 않도록 parted 패키지가 변경이 되었다고 합니다. (partprobe는 parted패키지에서 제공하는 유틸)

이유는 RHEL5를 비롯한 이전버전의 parted 패키지에서는 변경된 파티션 정보를 커널에 적용하는 과정에서 오류를 체크하는 부분이 없었기 때문입니다. partprobe가 종종 동작되지 않을 때도, 그 발생한 오류는 사용자에게 보고되지 않기 때문에 사용자는 파티션 작업이 정상적으로 적용되었다고 착각할 수 있지만, 실제 장치에서는 파티션 변경 내용이 적용되지 않아 데이터 입출력 간의 중첩쓰기로 인한 데이터 손실 등이 우려되므로, RHEL6에서는 데이터 안정성을 위해 사용중인 장치의 경우 변경된 파티션 테이블을 갱신하지 못하도록 수정되었습니다.

따라서 partprobe를 이용하여 변경된 파티션 정보를 다시 읽어 커널의 파티션테이블을 갱신하도록 하려면 mount 되어 있는 장치를 모두 umount 한 후 partprobe를 사용하거나 서버를 리부팅하여야 합니다.

하지만 partprobe를 사용하지 않고도 커널의 파티션테이블을 갱신할 수 있는 방법이 있는데,
partx 명령의 기능을 사용하면 사용중인 장치를 umount하거나 리부팅하지 않고도 커널의 파티션테이블을 갱신할 수 있습니다. (partx는 util-linux-ng 패키지에서 제공하는 유틸)

# partx -v -a /dev/sda

Issue

• Added a new partition to a disk that have some existing partitions already mounted. Can the new partition be used without reboot in RHEL6?

Environment

• Red Hat Enterprise Linux (RHEL) 6

Resolution

partprobe was commonly used in RHEL 5 to inform the OS of partition table changes on the disk. In RHEL 6, it will only trigger the OS to update the partitions on a disk that none of its partitions are in use (e.g. mounted). If any partition on a disk is in use, partprobe will not trigger the OS to update partitions in the system because it is considered unsafe in some situations.

So in general we would suggest:

1. Unmount all the partitions of the disk before modifying the partition table on the disk, and then runpartprobe to update the partitions in system.
2. If this is not possible (e.g. the mounted partition is a system partition), reboot the system after modifying the partition table. The partitions information will be re-read after reboot.

If a new partition was added and none of the existing partitions were modified, consider using the partxcommand to update the system partition table. Do note that the partx command does not do much checking between the new and the existing partition table in the system and assumes the user knows what they are are doing. So it can corrupt the data on disk if the existing partitions are modified or the partition table is not set correctly. So use at one's own risk.

For example, a partition #1 is an existing partition and a new partition #2 is already added in /dev/sdb by fdisk. Here we use partx -v -a /dev/sdb to add the new partition to the system:

# ls /dev/sdb*  
/dev/sdb  /dev/sdb1  

List the partition table of disk:

# partx -l /dev/sdb
# 1:        63-   505007 (   504945 sectors,    258 MB)  
# 2:    505008-  1010015 (   505008 sectors,    258 MB)  
# 4:         0-       -1 (        0 sectors,      0 MB)  
# 3:         0-       -1 (        0 sectors,      0 MB)  

Read disk and try to add all partitions to the system:

# partx -v -a /dev/sdb                                         
device /dev/sdb: start 0 size 2097152  
gpt: 0 slices  
ors,    258 MB)
dos: 4 slices  
# 1:        63-   505007 (   504945 sec
t  
# 2:    505008-  1010015 (   505008 sectors,    258 MB)  
         0-       -1 (        0 sectors,      0 MB)  
BLK
# 3:         0-       -1 (        0 sectors,      0 MB)  
# 4
:PG: Device or resource busy
error adding partition 1

(These last 2 lines are normal in this case because partition 1 is already added in the system before partition 2 is added)

Check that we have device nodes for /dev/sdb itself and the partitions on it:

# ls /dev/sdb*  
/dev/sdb  /dev/sdb1  /dev/sdb2

Comments

Ryan Shannon

August 8 2013 at 2:22 PMPermalink

The question is how to use a new partition without rebooting, and the solution is to take the disk out of service (equivalent to rebooting, from a downtime perspective) or reboot? Dynamic resizing of disks is very common in SAN and VMWare environments, and this "enhancement" to RHEL 6 that disables dynamic updates to active partition tables is ill advised, in my opinion.

gleb.shulman

August 16 2013 at 7:29 PMPermalink

The solution contains non-reboot way too using partx.

Just experimented with partitions merge:

- from /dev/sda

- removed partition 9

- removed partition 8

- recreated partition 8 twice big

- issued

partx -d -nr 8-9 /dev/sda

- issued

partx -a /dev/sda

- checked /proc/partitions

- kernel now is seeing /dev/sda8 only and reports new size.

 

If it will work without surprises, then would think that was an enhancement actually.

Partprobe became honestly advisable and 100% safe and split out its new risky "partx" twin-brother.

^IMO