Created attachment 1366458 [details] Detailed list of installed packages, smb.conf, smbd-debugger-output, smbd-strace-output, volinfo, etc. Description of problem: We found a strange behaviour with our gluster installation using Gluster 3.12 together with Samba (+Gluster-Vfs Plugin). Please find attached the detailed list of installed packages, smb.conf, smbd-debugger-output, smbd-strace-output, volinfo, etc. Writing to samba share stops if another gluster node becomes unavailable Version-Release number of selected component (if applicable): gluster3.12, samba4.6.2, samba-vfs-glusterfs4.6.2 How reproducible: We’re using a setup with two nodes. A volume with one brick on each node, Replica 2. Samba provides a share on Node1. Start writing a big file to the samba share on Node1 and disconnect Node2 (we stopped the VM). Actual results: Writing to samba share will freeze. It depends on the samba client if the write process will proceed after ~70 second (or longer) or if it fails Expected results: Writing to samba share will not stop or interrupted if another gluster node becomes unavailable Additional info: The attached files suggest the suspicion that the running gluster process is not aware in the fact that the other node died and is blocked until the dead node is available again or a timeout occurs. We also tried to use a fuse mount (samba used it as a local directory) with nearly the same result. Because of the long timeout the smb client might fail.
The behaviour above is caused by the "network.ping-timeout" default value of 42. Reduce this value if you are using a smb client. The Windows SMB Client use a timeout of 25. So 42 is too high for this Client.