What is role of bloom filter in cassandra?
From two different links of the Cassandra's documentation, I found:
A structure stored in memory that checks if row data exists in the memtable before accessing SSTables on disk
and
Cassandra checks the Bloom filter to discover which SSTables are likely to have the request partition data.
My question is does both the above statements are right? If yes, does bloom filters maintained for a Memtable and SSTable separately? Thanks in advance.
Answer
A Bloom filter is a generic data structure used to check if an element is present in a set or not. Its algorithm is designed to be extremely fast, at the cost of risking to return false positives.
Cassandra uses bloom filters to test if any of the SSTables is likely to contain the requested partition key or not, without actually having to read their contents (and thus avoiding expensive IO operations).
If a bloom filter returns false for a given partition key, then it is absolutely certain that the partition key is not present in the corresponding SSTable; if it returns true, however, then the SSTable is likely to contain the partition key. When this happens, Cassandra will resort to more sophisticated techniques to determine if it needs to read that SSTable or not. Note that bloom filters are consulted for most reads, and updated only during some writes (when a memtable is flushed to disk). You can read more about Cassandra's read path here.
Back to your questions:
1) The first statement ("A structure stored in memory that checks if row data exists in the memtable before accessing SSTables on disk") is IMHO not accurate: bloom filters are indeed updated when a memtable is flushed to disk, but they do not reference the memtable.
2) Bloom filters are maintained per SSTable, i.e. each SSTable on disk gets a corresponding bloom filter in memory.