Türker Yıldırım

indexing

  • Eleven indexes that existed only for reporting

    Note1 min read

    Indexes on production tables serving queries that no user request ever makes. An index that serves only reporting is paying a write cost on every insert to make…

  • Write latency after dropping an index

    Note2 min read

    Eleven indexes removed from production tables, and the write path measured before and after. The median saving is trivial and the tail is not, which is the usual…

  • The reporting database that is not a replica

    Deep dive10 min read

    A read replica serving reports, and reports that want a different schema. Eleven indexes on production tables that exist only for reporting.

  • Write latency, before and after eleven indexes

    Note1 min read

    Eleven reporting indexes dropped from production tables, measured on the write path. The median saving is trivial and the tail is not, which is the usual shape of…

  • A search index rebuilt because the mapping changed again

    Note1 min read

    A field that had to change type for the second time in eighteen months, and a rebuild that is now routine. The rebuild time is what makes a…

  • A schema change on a table nobody can lock

    Deep dive10 min read

    A column type change on a table taking four hundred writes a second, and an ALTER that estimates four hours.

  • The archive table nobody could delete from

    Deep dive10 min read

    900 million rows, a retention policy, and a legal question nobody had asked. A table that grows forever because deleting from it might be illegal.

  • A generated column that removed a JSON scan

    Note1 min read

    A key inside a JSON column filtered on four hundred thousand times a day, promoted without changing any application code. The optimiser matching a functional expression to a…

  • Two years of query logs, read once

    Deep dive10 min read

    A slow query log rotating for two years and never aggregated, and performance work driven by whoever complained loudest.

  • The endpoint that was slow for one customer

    Deep dive9 min read

    A healthy p95 and one customer for whom the endpoint takes four seconds. An aggregate metric that averages away the only case that matters.