Ask HN: Glusterfs or ceph on Digitalocean block storage?
I'm curious to hear of experiences running distributed filesystems like glusterfs or ceph (others too?) on Digitalocean's block storage.
I've got a side project that will be continuously adding many 50kb to 20mb files, so easy expansion is necessary.
Any "at-scale" (ie. production, under real load) performance reports/benchmarks appreciated.
Reliability, recovery from node failure, etc comments welcome. Don't use either. I tried and failed with both. Expensive hardware and a PhD required to go through all the config options. When I did get both to work, they were very slow and I kept getting replication issues and bricks that wouldn't heal. This cost two months of my life and I was not flavur of the month with my employer. I eventually tried lsyncd ( https://axkibe.github.io/lsyncd/) and it works like a charm. It took me about a week to get a decent setup going, but it has never let me down since, can handle large numbers of small files (similar to yours) and smaller numbers of large files. I use unison ( https://www.howtoforge.com/tutorial/unison-file-sync-between...) for web server pairs. Unless I'm misunderstanding, this isn't what I'm looking for. I'm looking for expandable storage - the ability to add GB/TB by adding disks. Replication is important, yes, but if I can't grow the volume, I'll just plain cry. You can add storage by expanding the lsyncd droplet(remote) or by using Block Storage(local) on your web droplets and expanding that. I have found this to be more cost effective than anything else I have tried. and I am also Digital Ocean customer. It has meant I can completely remove a whole storage layer and has reduced my droplet bill overall by 20%. Sorry for Unison, I should point out that I use fcron for running Unison every 5 seconds. This on average syncs 15-20 20Mb files every 5 seconds without any issue. This is similar to what you describe in terms of load. Don't. Please elaborate. Just "Don't" begs for more information.