Cortex vs OpenObserve

Cortex publishes its release notes as release candidates, and the two majors visible here tell a consistent story. 1.20 introduced Parquet-based block storage, Prometheus Remote Write 2.0, a resource-based limiter for ingesters and store-gateways, and regex tenant resolution, all experimental. 1.21 then made Parquet a Store Gateway query mode, graduated the Ruler API, Alertmanager API and sharding, tenant federation, FIFO/Redis cache and memcached service discovery out of experimental, and enabled the bucket index by default. The 1.21.1 candidate is almost entirely security hardening.

Two arcs run in parallel. The storage arc is moving Cortex off pure TSDB blocks toward Parquet, first as a converter and format, then as a queryable Store Gateway mode — the same columnar direction the rest of the observability market has taken for long-retention data. The maturity arc is a deliberate clearing of the experimental backlog, with six subsystems graduating in one release and their flags renamed to drop the experimental prefix. Layered on both is a distinct hardening push: the latest candidate bounds gossip connections, caps decompressed request bodies, adds HMAC stream authentication and validates tenant IDs against the authenticated caller.

Expect Parquet to keep progressing from experimental toward default, following the path the bucket index just completed. The security work in 1.21.1 reads like a systematic pass over untrusted input paths rather than a response to a single report, so more of the same in the ingestion and gossip layers is likely.