opencurve / curveadm Goto Github PK

I hope to upgrade the curvefs cluster that has been installed, but I have found that there are no relevant steps and effective upgrade methods.

[root@prit2-curve-kylin-node1 curvebs-deploy]# curveadm map wp:/vol3 --size 1000GB -c client.yaml  # wp:/vol3 is not exist
Start NEBD Service: [SKIP]
Map Volume: [OK]
[root@prit2-curve-kylin-node1 curvebs-deploy]# lsblk
NAME          MAJ:MIN RM   SIZE RO TYPE MOUNTPOINT
sda             8:0    0 446.6G  0 disk
├─sda1          8:1    0   300M  0 part /boot/efi
└─sda2          8:2    0 356.7G  0 part
  ├─klas-root 252:0    0 340.7G  0 lvm  /
  └─klas-tmp  252:1    0    16G  0 lvm  /tmp
sdb             8:16   0   1.8T  0 disk /mnt/minio-data
sdc             8:32   0   1.8T  0 disk /mnt/monitor
sdd             8:48   0   1.8T  0 disk
sde             8:64   0   1.8T  0 disk
sdf             8:80   0   1.8T  0 disk
sdg             8:96   0   1.8T  0 disk
sdh             8:112  0   1.8T  0 disk
sdi             8:128  0   1.8T  0 disk
sdj             8:144  0   1.8T  0 disk
sdk             8:160  0   1.8T  0 disk

[root@prit2-curve-kylin-node1 curveadm]# docker logs curvebs-target-daemon
tgtd: iser_ib_init(3421) Failed to initialize RDMA; load kernel modules?
tgtd: bs_curve_open(526) can not open curve volume cbd:pool//vol333_wp_, Success

if we add --create to map or target add command, it will mapped or added successfully.

Usage Example:

Run Plugin

• basic:

curveadm plugin run deploy-granafa -c config.yaml
curveadm plugin run mkfs-ext4 /dev/sda /dev/sbb /dev/sdc
curveadm plugin run shell --host "*" 'rm -rf /tmp/*' 
curveadm plugin run scp /path/to/local /path/to/remote

• alias:

CURVEADM_INVENTORY="~/hosts"
alias nessh='curveadm plugin run ssh'
nessh host1

• combine plugin:

# vim my-plugin.
curveadm plugin run plugin1
curveadm plugin run plugin2
curveadm plugin run plugin3

Add Plugin

curveadm plugin add PLUGIN-NAME -f /path/to/script
curveadm plugin add PLUGIN-NAME github.com/opencurve/curveadm/plugins/mkfs

List Plugin

curveadm plugin ls

Note:

• The plugin should support golang、bash、C++ and etc.
• The plugin build as one binary and store at ~/.curveadm/plugin.
• We need plugin storage and let user to contribute their plugin.

Few users may be unlike to deploy curve client with docker, we should handle this problem.
And if we deploy client for openstack or k8s csi, the client deployment with docker cannot be used by them.

when deploy curvefs cluster, will report error like this:

[2022/03/14 11:28:16.217 +08:00] [ERROR] [module.go:99] [execCommand] [[email protected]:22] [command="sudo docker exec 11d54334db424ae539d20bae721523f3471fe31ecc8ab89b583ecd33ef098ed8 bash /curvefs/tools/sbin/wait_chunkservers.sh"] [error="Process exited with status 1"] [output="/curvefs/tools/sbin/wait_chunkservers.sh: line 5: curve_ops_tool: command not found

At present, when cleaning services or clusters, the prompt information is too simple and not clear enough, and users may not be able to clearly understand the consequences of performing related operations, especially those that may cause service suspension or data loss operations. A stronger warning should be displayed to the user, and more complex operation commands to prevent users from misoperation.

[curve@localhost curveadm.new]$ curveadm clean
clean: role=* host=* id=* only=log,data,container
Do you want to continue? [y/N]: (default=N) y
Clean Service: [OK]


[curve@localhost curveadm.new]$ curveadm cluster rm first-cluster
Do you want to continue? [y/N]: (default=N) y
Deleted cluster 'first-cluster'

binaryPath can be empty in reload command if I just want to reload service.

We know that curveadm support many commands, it's a tedious work for user to type complete command every times, please support bash completion for curveadm, thx!

when use curveadm support command, we only collect server service logs now, we may need collect clients logs by given host and mountpoint.

You know that the trouble exists in the every corner of the world, we need help from professionals, please build a bridge between us and them, thx!

When I mount with the wrong yaml file, there is no error message.
Error when mount again:
Error: path mounted, please run 'curveadm umount ***' first,

But when using curveadm umount, there is an error, there is an error in the log:
output="***: not mounted

such as chunkserver disable ext4 renameat2 feature, nbd kernel module problem, nbd block size is 512bytes problem...

when create fs fail, but curveadm show ok

We use docker CLI to exec all managment operations currently, it's better to use API for compatibility, stability and safety.

When I want to update mds or metaser, I can only redeploy the entire cluster. It is recommended to add the function of upgrading the module separately.

currently we can deploy prometheus/grafana with docker-compose, but the target.json can not be generated and updated automaticly, we need to deploy them by curveadm and support generating and updating target.json.

Many functions of the client deployed with docker cannot be used:

1. Services such as cinder cannot be used
2. nbd's automount mount cannot be used
3. curve command does not work 。Including striping, recyclebin recovery volume, etc.
4. snapshot command does not work
5. curve-tool command does not work .Unable to set chunkserver pending
6. nbd map cannot specify parameters
7. 。。。

If users can not pull image from docker hub, then they can not format disks.

Curveadm status can list mds of curvefs. Can curveadm status print which mds is the leader mds?

such as pulling image or mounting dir.

mds/metaserver/space binary name has been changed since opencurve/curve#755, please check whether curveadm should make corresponding modifications.

The current operation of upgrading the cluster version, scaleout the cluster, the operation of upgrading the curveadm tool itself, and other scenarios that require a complex combination of steps, lack of corresponding guidance documents, and it is recommended that they can be improved.

Also we need add some descriptions for configurations, who deploy a cluster can refer these and know how to change them.

Both copies are on the same machine

• --no-recycle not work
• please support CurveBS v1.3, it will generate chunk file, like 3126.clean

We want to get the same results when checking /mnt/curve-fuse/ and /mnt/curve-fuse.

• install script

sh -c "$(curl -fsSL https://curveadm.nos-eastchina1.126.net/script/install.sh)"

install.sh: 14: Syntax error: "(" unexpected
install.sh: 110: [: ==: unexpected operator

The commands in the code are now prefixed with sudo by default, which may introduce

1. security problems
2. users actually have permission to follow commands, but not a sudoer, causing tasks to fail. (For example, the user is in the docker group, but not a sudoer)

I feel that sudo should be included as an optional parameter -sudo or --root ? and only add sudo in front of the command after enabling the parameter.

this issue related to golang/go#38373

make failed on

OS: CentOS Linux release 7.9.2009 (Core)
golang version: go version go1.16.13 linux/amd64
git version: git version 1.8.3.1

error messages(an example, may be different):

# make
go build -o sbin/curveadm /root/curveadm/cmd/curveadm/main.go
go: github.com/theupdateframework/[email protected] requires
        github.com/docker/[email protected]: invalid pseudo-version: git fetch --unshallow -f origin in /root/go/pkg/mod/cache/vcs/48fbd2dfabec81f4c93170677bfc89087d4bec07a2d08f6ca5ce3d17962677ee: exit status 128:
        fatal: git fetch-pack: expected shallow list
make: *** [build] Error 1

make successfully after upgrading git version to:
git version: git version 2.24.4

yum -y erase git # this command will remove golang too, so need to install golang again
yum -y install https://repo.ius.io/ius-release-el7.rpm
yum -y install git224 golang

All operation commands need to be recorded to the log file for follow-up audit operation steps.

related issue: opencurve/curve#939

1. When mounting, the format of mountpoint should be: [hostname:mountpoint], but when using curveadm mount, the format of mountpoint becomes [a string of numbers: /usr/local/curvefs/client/mnt]
2. Although unmounting with umount is successful, the mount point is unavailable and an error output appears.
   
   
3. Although it shows that umount is successfully uninstalled, the mount point is still inaccessible, and the error output is as follows:
   
4. At this time, deleting fs will fail, mds log shows fs_busy, you need to use the tool umount, you can delete successfully at this time.

current prompt message during run status command:

[curve@localhost curveadm]$ curveadm cluster add xxx
[curve@localhost curveadm]$ curveadm status
Error: unsupport kind('')

[curve@localhost curveadm]$ curveadm cluster ls
  xxx
[curve@localhost curveadm]$ curveadm cluster checkout xxx
Switched to cluster 'xxx'
[curve@localhost curveadm]$ curveadm status
Error: unsupport kind('')

[curve@localhost curveadm]$ curveadm cluster add yyy -f ../curveadm.new/noservice.yaml
[curve@localhost curveadm]$ curveadm status
Error: unsupport kind('')
[curve@localhost curveadm]$ curveadm cluster checkout yyy
Switched to cluster 'yyy'
[curve@localhost curveadm]$ curveadm status
No service[curve@localhost curveadm]$

[curve@localhost curveadm]$ ./sbin/curveadm cluster add xxx -f ../curveadm.new/newtop.yaml
[curve@localhost curveadm]$ curveadm cluster checkout xxx
Switched to cluster 'xxx'
[curve@localhost curveadm]$ curveadm status
Error:
service(id=71687ea52675) not found
See log file for deatil: /home/curve/.curveadm/logs/curveadm-2022-01-28_01-15-16.log    ### this log file is not exist.

we may change them to these:

[curve@localhost curveadm]$ ./sbin/curveadm cluster add xxx
[curve@localhost curveadm]$ ./sbin/curveadm status
No cluster, please add a cluster firstly
[curve@localhost curveadm]$ ./sbin/curveadm cluster checkout xxx
Switched to cluster 'xxx'
[curve@localhost curveadm]$ ./sbin/curveadm status
Error: cluster topology is empty

[curve@localhost curveadm]$ ./sbin/curveadm cluster add xxx -f ../curveadm.new/noservice.yaml
[curve@localhost curveadm]$ ./sbin/curveadm cluster checkout xxx
Switched to cluster 'xxx'
[curve@localhost curveadm]$ ./sbin/curveadm status
No service, please check your cluster topology

[curve@localhost curveadm]$ ./sbin/curveadm cluster add xxx -f ../curveadm.new/newtop.yaml
[curve@localhost curveadm]$ ./sbin/curveadm status
No cluster, please add a cluster firstly
[curve@localhost curveadm]$ ./sbin/curveadm cluster checkout xxx
Switched to cluster 'xxx'
[curve@localhost curveadm]$ ./sbin/curveadm status
Error:
service(id=d585d74ac18f) not found
Did you deploy the cluster?

We hope to know how many users are using curve and how large it is. Therefore, we can provide a reporting service to obtain information about the scale of user clusters on a regular basis, and perform statistics and display of relevant information on the server.

deploy curvebs stuck in Create Physical Pool,

beause it is stuck in curl second mds, while major mds is third mds.

see https://github1s.com/opencurve/curve/blob/HEAD/curve-ansible/common_tasks/check_chunkserver.yml

$ curveadm map <user>:<volume-name> -c client.yaml --create --size 10GB
<volume-name> must be start with '/' , but many users ignored it even we describe this note in manual.

so I think we should add some checks about the format of <volume-name>, such as if it is not start with '/', we print some error messages to users to suggest them using correct format.