bob
bob copied to clipboard
qoollo
Distributed BLOB storage
In the current implementation first nodes in the list are chosen for every request. As a result, these nodes contain most of the aliens. We should achieve an uniform distribution...
Current implementation does not check nodes availability status and therefore leads to a decrease in efficiency and reliability. https://github.com/qoollo/bob/blob/eab98a3cd1614b1eae1df20809559bbfb3f20573/bob/src/cluster/quorum.rs#L142 https://github.com/qoollo/bob/blob/ab2b97c32fed59fe31aad50c1610ab3891777a5c/bob-common/src/mapper.rs#L118
Memory metrics: bob_ram: 29 Gb bloom_filter_ram: 10.5 Gb index_memory: 15 Gb Index_memory looks suspicious, because `open_blobs_hard_limit` is 10
This update requires changes in the source code. Related issue: https://github.com/qoollo/bob/issues/551
Currently, we authorize nodes by pair `(node_name, ip_address)`. In some network configurations it is not possible to correctly determine the IP address, so we need an alternative. We can add...
Currently, if node is not available during deletion, the data on it would not be marked as deleted even after its recovery. In this case, the data in the cluster...
At the moment, all logic is in Pearl, which is not good: BLOB v1 has no changes in structures compared to BLOB v0 (only data has changed). We should implement...
Currently, Pearl `Key` wraps `Vec` (https://github.com/qoollo/bob/blob/master/bob-backend/src/pearl/data.rs#L6), which is inefficient for key sizes less then or equal to 16. At the same time we have `BobKey` struct (https://github.com/qoollo/bob/blob/master/bob-common/src/data.rs#L15) that already wraps...
https://github.com/qoollo/bob/blob/master/bob/src/cluster/quorum.rs#L212