GreptimeTeam/greptimedb

What type of bug is this? User Experience What subsystems are affected? Other Minimal reproduce step greptime cli meta snapshot save --backend postgres-store \ --store-addrs 'password=... dbname=postgres user=postgres host=localhost port=5432' \ --meta-schema-name greptime_schema \ --auto-create-schema false What did you expect to see? The CLI accepts a way to turn off automatic schema creation, as Metasrv does with auto_create_schema = false. What did you see instead? error: unexpected argument 'false' found --auto-create-schema is a bool field with default_value_t = true: greptimedb/src/cli/src/common/store.rs Lines 87 to 90 in 3a4c24a /// Automatically create PostgreSQL schema if it doesn't exist (default: true). #[cfg(feature = "pg_kvbackend")] #[clap(long, default_value_t = true)]

What type of bug is this? Incorrect result What subsystems are affected? Write Protocols, Frontend Minimal reproduce step Upload two versions of the same pipeline. Version 1 writes name into column name_v1; version 2 writes it into name_v2: version: 2 processors: - date: field: ts formats: - "%Y-%m-%dT%H:%M:%S%.fZ" transform: - field: name, name_v1 # name_v2 in the second version type: string - field: ts type: time index: timestamp Write through _bulk, selecting version 1 with the header, and separately with the URL parameter: V1='<version string returned for the first upload>' curl -X POST 'localhost:4000/v1/elasticsearch/_bulk?pipeline_name=esv' \ -H 'Content-Type: application/json' \ -H "x-greptime-pipeline-version: $V1" \ --data-binary $'{"create":{"_index":"es_hdr"}}\n{"name":"alice","ts":"2024-01-01T00:00:00.000Z"}\n' curl -X POST "localhost:4000/v1/elasticsearch/_bulk?pipeline_name=esv&version=<url-encoded $V1>" \ -H 'Content-Typ

What type of enhancement is this? Refactor What does the enhancement do? Five aggregate functions in common-function expose their AggregateUDF through a method misspelled as uadf_impl. The other aggregates in the same module use udf_impl (CountHash::udf_impl(), AvgAccumulator::state_udf_impl() / merge_udf_impl()), so rename these to udf_impl too. Occurrences on main (10 in total): src/common/function/src/aggrs/vector/sum.rs, vector/product.rs, vector/avg.rs: the definitions on VectorSum, VectorProduct, VectorAvg src/common/function/src/aggrs/vector.rs: 3 callers src/common/function/src/aggrs/geo/geo_path.rs, geo/encoding.rs: the definitions on GeoPathAccumulator, JsonEncodePathAccumulator src/common/function/src/aggrs/geo.rs: 2 callers git grep uadf_impl should return nothing afterwards, and cargo check -p common-function should pass. No behavior change. The vector files are also touched by #9338; if it's still open, rebase after it merges. Implementation challenges No response

What type of enhancement is this? API improvement, Configuration What does the enhancement do? gRPC-Web was added to the frontend gRPC server in #8027, but browsers still can't call it cross-origin because there is no CORS handling. tonic_web::GrpcWebLayer (0.14) doesn't answer preflights itself; its docs expect a tower_http::cors::CorsLayer on top. Repro on v1.2.1: # gRPC-Web itself works $ printf '\x00\x00\x00\x00\x00' | curl -s --http1.1 -i -X POST \ http://127.0.0.1:4001/greptime.v1.HealthCheck/HealthCheck \ -H 'content-type: application/grpc-web+proto' -H 'x-grpc-web: 1' \ -H 'Origin: http://example.com' --data-binary @- HTTP/1.1 200 OK content-type: application/grpc-web+proto ... # the browser preflight fails $ curl -s --http1.1 -i -X OPTIONS \ http://127.0.0.1:4001/greptime.v1.GreptimeDatabase/Handle \ -H 'Origin: http://example.com' \ -H 'Access-Control-Request-Method: POST' \ -H 'Access-Control-Request-Headers: content-type,x-grpc-web' HTTP/1.1 400

up == 0 and node_load1 > 5 come back without __name__. Prometheus keeps the left-hand metric name for comparison operators; only arithmetic operators and bool comparisons drop it. Grafana legends for those panels render as {instance="..."} instead of node_load1{instance="..."}, and alert templates that read {{ $labels.__name__ }} get nothing. Where GreptimeDB stores the metric name as the table name, so the __name__ of a response is derived from the query AST, not from the data. That happens in collect_metric_names in src/servers/src/http/prometheus.rs: PromqlExpr::Binary(BinaryExpr { lhs, op, .. }) => { if matches!(op.id(), token::T_LAND | token::T_LOR | token::T_LUNLESS) { collect_metric_names(lhs, metric_names) } else { metric_names.clear() } } Every binary operator that is not and/or/unless clears the name. Comparison operators should not. The rule to implement Prometheus drops the metric name when shouldDropMetricName(op) holds — + - * / % ^ atan2 —