Request testing feedback for component/feature bitrot in release-3.10 about glusterfs HOT 2 CLOSED

Testing the bitrot with Brick multiplexing enabled. Performed following tests.

• Created a replicate volume and started the volume. There was only 1 process for all the bricks on the node (Created another volume and started it to ensure)

• Mounted the volume and created some data.

• Enabled bitrot on the replicate volume.

• BitD was able to sign the already existing files on the volume

• Enabled ondemand scrubbing and scrubber completed in the default lazy mode

• Modified one of the files from the backend directly to simulate backend corruption of the file.

• Agin triggered on demand scrubbing. Found that the scrubber has marked the file which was modified directly in the backend as a bad file via extended attribute.

• Deleted the bad file from the Backend (along with the gfid handle) and triggered a heal from the mount point by accessing the file.

• Found that the good file from the other replica copy has been properly healed to the other node where bad copy was present.

TODO:

• Create more data and see the freshly created data after bitrot enable also gets signed properly or not.

• Try different throttling options of scrubbing

• Try accessing a bad object from the mount point and check if error is generated or not.

• Before the object gets signed again modify it and check if the object gets the proper signature or not

from glusterfs.

Did few more testings.

• Created new data on a bitrot enabled volume and saw that the newly created files are getting signed.

• The signer (i.e. BitD) waits for 2 minutes once it gets a notification that a file has changed, before signing the file. There can be a race window where within the 2 minutes wait period, the file again gets modified and a new notification is sent to BitD. In this case BitD should discard previous notification and freshly start signing after 2 minutes. Verified this by modifying a file immediately after closing it.

TODO:

• Try different throttling options of scrubbing

• Try accessing a bad object from the mount point and check if error is generated or not.

from glusterfs.