ObservabilityEngineering30 September 20263 min read

The in-memory observability database learns the SQL its dashboards need

The in-memory database behind the observability catalog gains string functions, searched CASE, arithmetic, CAST and typed computed columns, and its columns keep labels, content types and colours through every SELECT.

The observability catalog is served by an in-memory database that answers SQL over live JVM state. Its dialect was narrow: a CASE threw on anything beyond AND-ed comparisons, ORDER BY accepted only column names, and every computed column came back as text. Formatting a dashboard meant writing a script. This release gives the engine the vocabulary and lets a column's declarations travel with its data.

Expressions

String functions UPPER, LOWER, TRIM in its three forms, LENGTH, SUBSTR, REPLACE, CONCAT and ||, LPAD, RPAD, INSTR and LOCATE, in the select list, the WHERE and the ORDER BY. NULL in, NULL out. They nest and sit inside CASE.
Searched CASE with the full WHERE vocabulary OR, NOT, IS NULL, LIKE, IN, BETWEEN and parentheses, anywhere an expression goes. ORDER BY evaluates expressions and accepts positions.
Arithmetic and CAST The four operators and modulo with precedence, ROUND, TRUNC, MOD and ABS, and CAST to the integer, decimal, double, varchar, timestamp, date and boolean types. A timestamp accepts epoch milliseconds, and division by zero yields NULL.
Typed results A computed column takes the type of what it reads: arithmetic promotes, CASE merges its branch types, COALESCE follows its first non-null. Grids render durations, percentages and instants as such instead of as text.

Declarations that survive a query

Column declarations A column declares its label, header group, tooltip, render options and content type, and a SELECT clones them onto the projection. An alias renames the clone.
Content type from the row A content type may name a column of the row, so a scripts table mixing JavaScript and Python opens each run in the right editor.
State colours A state column declares its vocabulary: green running, blue waiting, yellow attention, red failed, grey inert, white text throughout. Kinds that carry no judgement, a thread type or a replica role, get categorical hues that never read as good or bad.
Percent heat maps A seven-step ramp from green through amber to red, with white text at a contrast of at least 3:1 at every step. A null percent is unshaded and a caption in a percent column is left alone.

Privileges and DML

Privileges are the table's own and bind every user, the DBA included. DELETE is offered only when a trigger exists to act on it, so a delete on a snapshot never silently does nothing.
No transactions to manage Row DML runs without begin, commit or rollback. Every statement is atomic.

The payoff is on the other side of the query: every provider's tables, JVM, JDBC, statistics, cache and machine, now read at a glance with short labels under themed headers, and the dashboards that depended on scripts for their formatting no longer need them.