Storage: Add Pure Storage storage driver #14599
Conversation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
MusicDin
force-pushed
the
feat/pure-driver
branch
2 times, most recently
from
c53014e to
69f6cae
Compare
doc/howto/storage_pools.md
Outdated
|
|
||
| Create a storage pool named `pool1` that uses NVMe/TCP by default: | ||
|
|
||
| lxc storage create pool1 pure pure.gateway=https://<purestorage-address> pure.api.token=<purestorage-api-token> pure.array.address=<nvme-ip> |
There was a problem hiding this comment.
That gateway HTTPS verification presumably relies on the OS provided trust store so it would work if it's an "official TLS" cert like one signed by Let's Encrypt.
I must admit not knowing how to provide a different trusted CA if one doesn't have an official cert.
MusicDin
force-pushed
the
feat/pure-driver
branch
from
69f6cae to
231d0f0
Compare
MusicDin
force-pushed
the
feat/pure-driver
branch
from
231d0f0 to
276bb1b
Compare
|
@minaelee Thanks for the comments. I've addressed all of them, but will ask you for another review just before we merge the PR, as there may still be some changes to the docs. Thanks again :) |
Signed-off-by: Din Music <[email protected]>
…ge pool is created Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
…optimized image Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
… token are set Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
Correctly initialize Pure Storage pool. Signed-off-by: Din Music <[email protected]>
…lable storage drivers Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
Currently, Pure Storage driver does not support recovery. Mainly bacuse the storage volume names are encoded, which would result in indistinguishable storage volume names after the recovery. Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
… drivers Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
Signed-off-by: Din Music <[email protected]>
MusicDin
force-pushed
the
feat/pure-driver
branch
from
ac9c57b to
40b6b51
Compare
| return err | ||
| } | ||
|
|
||
| // Ensure temporary snapshot volume is remooved in case of an error. |
There was a problem hiding this comment.
not a blocker to merge, but typo here
|
|
||
| // A Pure Storage snapshot cannot be mounted. To mount a snapshot, a new volume | ||
| // has to be created from the snapshot. | ||
| err = d.client().copyVolumeSnapshot(snapVol.pool, parentVolName, snapVolName, snapVol.pool, snapVolName) |
There was a problem hiding this comment.
@MusicDin question for you:
What happens if I've got a snapshot mounted (so a temporary volume has been created), and then lxd gets stopped and the host restarted uncleanly, and i go to mount the snapshot again, will this call fail because the temporary volume already exists?
There was a problem hiding this comment.
copyVolumeSnapshot creates a new volume, or overwrites an existing one if it already exists.
Mounting is handled by the NVMe/iSCSI if host is mapped with a certain volume.
When we attempt to create a mapping, the error is ignored if it already exists, so it cannot fail there.
The only potential issue I can see is that if LXD is abruptly shut down, it will not cleanup the temporary snapshot. For example, if the instance is then removed, this temporary snapshot could remain in Pure Storage.
There was a problem hiding this comment.
^ Until the pool itself is removed, then everything is wiped.
There was a problem hiding this comment.
or overwrites an existing one if it already exists.
OK cool,that was the bit I was mainly interested in.
For example, if the instance is then removed, this temporary snapshot could remain in Pure Storage.
Can we scan volumes on instance (or snapshot) delete to find any temporary volumes related and delete them too?
There was a problem hiding this comment.
Can we scan volumes on instance (or snapshot) delete to find any temporary volumes related and delete them too?
I cannot foreseen anything that would block us from doing that.
What is your suggestion, at which phase should this happen? Should we run the cleanup on LXD boot / when storage is loaded?
There was a problem hiding this comment.
when deleting the instance snapshot, and when deleting the instance (which by extension is deleting the snapshot)
There was a problem hiding this comment.
Ahh yeah of course. So in practice this would mean when we are deleting a specific snapshot, we check if a volume exists with such uuid.
For example:
- Delete snapshot v1/snap0
- Check if there is an existing volume with
snap0-uuid(obviously correctly resolving volume prefixes and suffixes in the process) - Delete such volume if exists
- Delete an actual snapshot
snap0
|
|
||
| Create a storage pool named `pool4` that uses NVMe/TCP to connect to Pure Storage array via specific target addresses: | ||
|
|
||
| lxc storage create pool4 pure pure.gateway=https://<pure-storage-address> pure.api.token=<pure-storage-api-token> pure.mode=iscsi pure.target=<target_address_1>,<target_address_2> |
There was a problem hiding this comment.
uses NVMe/TCP to connect
pure.mode=iscsi
copy pasto? Please can you follow up in separate PR these small fixes.
Pure Storage does not allow mounting snapshots directly. Instead, we create a temporary volume with the snapshot's volume name. If the LXD was abruptly closed when temporary volume is being used, it may remain in the Pure Storage until the pool is removed. To prevent such situation, when snapshot is removed also remove the potentially existing temporary snapshot volume. Addresses comment: #14599 (comment)
Closes #14814 This aligns the PowerFlex driver with the implementation of Pure in #14599 to use the NVMe/TCP `Connect` and `Disconnect` functions instead of their `(Connect|Disconnect)All` counterparts. Furthermore a patch is introduced to remove the `powerflex.sdt` config key from already existing storage pools. Currently only a single address is stored in the key and using `nvme connect-all` all of the other targets (SDTs) are discovered anyway. As the driver is now querying for all the SDTs in case `powerflex.sdt` config key is not set, the behavior is the exact same. In addition `powerflex.sdt` now accepts multiple addresses separated by comma which allows overriding the addresses discovered by asking the PowerFlex REST API. Some additional minor tweaks have been put into place to make more consistent use of the connector's `LoadModules` in case of SDC and to clean up some error messages.
This PR adds
purestorage driver for Pure Storage remote storage.Driver supports communication over either iSCSI or NVMe/TCP.
iscsiadmto the LXD host: commands/daemon: Redirect iscsiadm to the host lxd-pkg-snap#646