The base workload is git checkout of random commits. In parallel there's read-only snapshot every second. The overall state of the space in the filesystem is below limits so a metadata operation should succeed. Stats saved after the error happens: there's some remaining metadata space (so snapshot should work), there is usable data space, global reserve is small but should be large enough(?). The whole disk is 2G, maybe the global reserve is not scaled properly(?).
The base workload is git checkout of random commits. In parallel there's read-only snapshot every second. The overall state of the space in the filesystem is below limits so a metadata operation should succeed. Stats saved after the error happens: there's some remaining metadata space (so snapshot should work), there is usable data space, global reserve is small but should be large enough(?). The whole disk is 2G, maybe the global reserve is not scaled properly(?).