There are 2 kinds of healing functions in GlusterFS, server-side heal and client-side heal.

Server side heal is automatically executed by self-heal daemon on all gluster server nodes. It does healing by crawling file/directory information from .glusterfs directory on brick path. So it will keep the file-data and meta-data to be consistent from server side.

Client side heal is different, it will triggers heal for the particular file whenever client accesses files from mount path, which means a file operation on file descriptor. It will lead to some performance impact as every file access operation will go through extra set of function calls for file check and heal. Client-side heal is enable by default, and can be turned off by below command

File data

# gluster volume set <volume name> off

Entry data (contents/entries of a directory)

# gluster volume set <volume name> cluster.entry-self-heal off

Meta data

# gluster volume set <volume name> cluster.metadata-self-heal off

Note that turn off client side healing doesn't mean to compromise data integrity and consistency. For file read pending xattr is evaluated for replica copies and read will only served from correct copy. For file write new data will be written on both replica bricks, and self-heal daemon on gluster node will take care of these and will do healing if needed.