DM-Multipath document says that multipathd support active/active clusters with some policies such as round-robin. However, there is some racing problem in this kind of clusters.

For example, assume there is a client who is connected to a disk in SAN storage through two separate SAN switches. If the client writes X and Y in the same block of disk through two different paths, which value will be written to the disk? How does DM-Multipath handle this racing problem?



The outcome of your scenario has nothing to do with multipathing, no production file system would allow for write-commit confirmations to get out of sync, whether a file system uses multipathing is irrelevant.

  • What will happen if I use two separate and parallel "dd" command to write the data on the disk? in this scenario, there is no file system – Gozal Jul 12 '16 at 12:35
  • DD just writes blocks, there's no coordination between multiple executions of it, if you had two of these and somehow made them out of sync they'd happily muddle the data up as you described - but a filesystem wouldn't, they maintain consistency via many methods including in-order block overwriting. – Chopper3 Jul 12 '16 at 14:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.