This one is already used on Satellites, but planning to start a new Storagenode rollout, and didn’t find related thread. Please note that the opt-out from hashstore is disabled with this release. Based on monitoring old piecestore data is less than 1 PB. We would like to finish hashstore migration …

Worst release ever, stupid idea to force cutoff piecestore for all.

Maintaining it wouldn’t work. With hashstore we can run a full audit of a node with a cheap hashtable lookup. That is impossible with piecestore. That is the main reason we want to move forward. From the satellite point of view it is better to run all nodes on hashstore even if that means we might h…

I didn’t know about this feature :person_shrugging:

[image] Alexey: Do you want to maintain it? For free? Absolutely no, because Storj is commercial company.

[image] littleskunk: With hashstore we can run a full audit of a node with a cheap hashtable lookup Can you elaborate? Without reading all data you can’t know it’s healthy regardless of the arrangement it’s stored in. [image] d4rk4: [image] Alexey: Do you want to maintain it? For …

[image] arrogantrabbit: Can you elaborate? Without reading all data you can’t know it’s healthy regardless of the arrangement it’s stored in. A bitflip would still be possible yes but a lot of bugs like lets say a piece that gets deleted by garbage collection would get visible almost for free…

[image] littleskunk: A bitflip would still be possible yes Or lost piece, or deleted piece [image] littleskunk: full audit I see. So this is not a “full audit” then, but rather a check, that can quickly detect some limited class of problems, but cannot actually confirm that data exist…

[image] arrogantrabbit: Or lost piece, or deleted piece I don’t see how that would be possible. How would we end up with a hashtable entry to a not existing piece? The data has to be written into a log file so it must be there and the only way to purge it is compact. [image] arrogantrabbi…

[image] littleskunk: he data has to be written into a log file so it must be there Nope. Writing a piece successfully does not guarantee it is still there later. A file can lose data while the hashtable still says the piece exists. It can disappear for many innocent (like bug in truncation/c…

Release preparation v1.164

Product Discussions

Alexey October 13, 2026, 4:07am 42

Seems more than that, some reported here that the compaction cannot pass for some logs: