Granular entry self-healing

A system comprising:

• a memory;
• and
• a processing device operatively coupled to the memory, the processing device to:
• receive a request to perform an operation on a file within a directory, the distributed file system comprising a first replica stored on a first server in a distributed file system and a second replica stored on a second server in the distributed file system, wherein the first replica and the second replica are both replicas of the directory;
• generate a first entry and a second entry in a file index, the first entry corresponding to the first replica and the second entry corresponding to the second replica, wherein the first entry and the second entry indicate the file within the first replica and the second replica on which the operation was requested to be performed, wherein the file index comprises a plurality of entries, and wherein the plurality of entries corresponds to replicas of directories stored on the second server that were modified during a period of time when the first server was out of service and are not added to the file index stored on the second server until the second server receives a notification that the first server is out of service;
• perform the operation on the first replica and on the second replica;
• and
• maintain the first entry in the file index responsive to not receiving a confirmation that the operation was successfully performed on the first replica.

A non-transitory machine-readable storage medium storing instructions which, when executed, cause a processing device to:

• determine, by the processing device, that a first entry corresponding to a first replica stored on a first server in a distributed file system is present in a file index, the first entry identifying a file within the first replica that was modified, wherein the file index comprises a plurality entries, and wherein the plurality of entries corresponds to replicas stored on a second server in the distributed file system that were modified during a period of time when the first server was out of service and are not added to the file index until the second server receives a notification that the first server is out of service;
• receive, by the first server, a copy of a second replica stored on the second server, wherein the first replica and the second replica are both replicas of a same directory;
• and
• modify the first replica in view of the second replica so that the first replica matches the second replica after an operation was performed on the second replica.