On Fri, Aug 25, 2023 at 3:23 PM Ryan Blue <b...@tabular.io> wrote: > I don't think that we should introduce nanosecond precision types without > at least supporting both timestamp and timestamptz. I'm not sure whether > nanosecond time should be supported. >
SGTM; this seems to be the most agreeable part of the proposal. For the primary keys, what is the use case you're trying to solve? Do your > tables allow null values in primary keys? If so, what is the purpose of it? > InfluxDB is a schema-on-write database; tables and columns are created by writing to them. Constraints: - Every table has exactly one timestamp[nanos] column, and is required. - "Field" columns are typed (int, uint, float, string, bool). These are the time series data that vary with time. At least one field value is required, per row. - "Tag" columns are only strings. These are identifying data - used for grouping, filtering. Tag values are not required, whether tag columns are present or not. Primary keys are composed of **non-null tags**, plus timestamp. For example, these rows: timestamp | tag A | tag B | field(int) F 09:25 | null | null | 1 09:25 | "foo" | null | 1 09:25 | "foo" | "bar" | 1 10:25 | "foo" | "bar" | 1 have these primary keys: (09:25) (09:25,A="foo") (09:25,A="foo",B="bar") (10:25,A="foo",B="bar") InfluxDB uses these primary keys in two contexts: - deduplication in query pipelines - compaction (mitigate performance impact of query-time deduplication) -- Jacob Marble 🇺🇸 🇺🇦
