You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Split metadata parsing into Scylla and Cassandra implementations #1247
Follow-up to #1159 and the metadata-discovery changes in #1242.
Problem
Metadata parsing shares a version-based schema parser hierarchy, while product-specific metadata query selection also lives in ControlConnection. The schema parser selector selects a parser by release major version; the local metadata query selector separately checks server capabilities and chooses columns.
The port-discovery fix illustrates the distinction: Cassandra 4+ exposes local broadcast/RPC port columns that Scylla does not provide. Supporting both requires compatible bootstrap queries, an optional column probe, and fallback handling. Product-specific rules are therefore spread across query construction and metadata parsing.
Expected behavior
Introduce separate Scylla and Cassandra metadata parser implementations behind a common interface, sharing common parsing machinery where appropriate. Each implementation should own its product-specific system-table queries, supported columns, and interpretation.
Define server/product selection and a safe fallback when identity or capabilities are incomplete; a release major version alone should not establish that the server is Cassandra.
Preserve public APIs, supported server compatibility, and the behavior covered by #1159 and #1242, including token-metadata options, parallel peers/local refresh requests, and connection-scoped fallback for unsupported columns. Cover parser selection, query projections, and metadata discovery/refresh with unit tests and appropriate Scylla/Cassandra integration tests.
Activity