Skip to content

PeachQ 0.92: partitioned databases load

The largest gap we listed at 0.91 was on-disk storage: partitioned and segmented databases did not load. In PeachQ 0.92 they do. \l mounts a partitioned or segmented database, and .Q.dpft and its relatives write partitions in the kdb+ layout.

What's new

  • Partitioned and segmented databases: \l mounts them and qSQL queries them, with the virtual columns date and i. .Q.dpft, .Q.dpfts, .Q.dpt, .Q.dpts and .Q.hdpf write partitions, and .Q.bv, .Q.vp and .Q.chk handle tables missing from a partition.
  • Splayed tables written in place: upsert and insert on a splay directory append to each column file, all or none, and @[`:dir/;`col;…] amends one column on disk.
  • S3 and compressed containers: s3:// reads natively with AWS SigV4 signing and per-bucket .s3 configuration. gzip, tar and zip files are recognised by their content and read through any transport.
  • TLS IPC: hopen takes tcps:// addresses, and a failed hopen names its cause.
  • Restricted and blocked modes: -u 1, -U and -b follow kdb+.
  • More q projects run: 14 of the 32 open-source projects on the compatibility page now fully work, up from 11 at 0.91, and 3 are broken, down from 6.
  • q conformance reaches 92.1% (91.4% at the last 0.91 build), and Windows is at or above its target on all 1294 measured suites.

An HDB in five lines

Write two days of trades with .Q.dpft, load the database and query across both partitions:

q)trade:([]sym:`IBM`MSFT`IBM;price:101.5 330.2 102.0;size:100 200 300)
q).Q.dpft[`:hdb;2026.11.13;`sym;`trade]
`trade
q)trade:([]sym:`MSFT`IBM;price:331.0 103.5;size:150 50)
q).Q.dpft[`:hdb;2026.11.14;`sym;`trade]
`trade
q)\l hdb
q)select vwap:size wavg price, volume:sum size by date,sym from trade
| date       | sym    | vwap    | volume |
| date       | symbol | float   | long   |
|============|========|---------|--------|
| 2026.11.13 | IBM    | 101.875 | 400    |
| 2026.11.13 | MSFT   | 330.2   | 200    |
| 2026.11.14 | IBM    | 103.5   | 50     |
| 2026.11.14 | MSFT   | 331     | 150    |

On disk this is the layout a kdb+ user expects: a sym file at the root, a directory per date, and a splayed trade table in each.

Correct first, fast next

Queries on a partitioned database are correct but not yet efficient. Each query builds one in-memory table from the partitions it reads, so a large HDB will be much slower than kdb+ and use far more memory. Performance, multi-threaded query execution, and snappy-compressed and encrypted files are the main gaps now; the roadmap lists them.

If your firm runs kdb+ in production and would like to try PeachQ against a real HDB, talk to us. Real workloads decide what we build next.

Download

Download 0.92 from the download page. The full list of changes is in the release notes.