Skip to content

Limits

Every size limit in Torment, and why each exists.

Limit Value Over the limit Why
Postings per request 8,189 on a production TigerBeetle cluster; about 250 on a --development one, like the local setup 400 invalid_request over 8,189, 400 batch_too_large over the cluster’s own limit TigerBeetle’s batch size, set by the cluster
Request body 1 MiB (about 5,000 postings) 400 protects the server from huge bodies. Can be raised if a caller needs it
Request time 5 s to send headers, 30 s to send the whole request, 30 s to receive the response the connection is closed protects the server from slow clients. A cut-off POST /transfers is a request with an unknown outcome: resend it with the same ids
Waiting on Postgres and TigerBeetle 10 s per request 500 a stalled database doesn’t hold requests open forever. For POST /transfers the outcome is unknown: resend with the same ids
Address length 255 bytes 400
Currency precision 0 to 18 decimal places 400
Kind name lowercase letters, digits and _, starting with a letter, at most 63 bytes 400
Kinds 65,535 500 on registering one more each kind is a TigerBeetle transfer code, a 16-bit number
Ids, group ids any UUID except the nil and max UUIDs 400 TigerBeetle reserves the 128-bit ids 0 and 2¹²⁸−1
Amounts 1 to 2¹²⁸−1 minor units 400
Page size (limit) default 100, from 1 to 1,000 400
Wallets in a pattern balance query 1,000 400, asking you to narrow the pattern each read covers every wallet at once
Wallets in a pattern history query 100 400, asking you to narrow the pattern each wallet is one query to TigerBeetle

Lists don’t have a total limit, only a page size. GET /accounts, history queries and group reads return next when more results remain: pass it back as cursor. Each page costs the same, however deep you go.

  • Accounts, wallets and currencies. As many as Postgres and TigerBeetle hold.
  • Postings over time. TigerBeetle is built to store billions of transfers.
  • Concurrent requests. Torment merges concurrent writes into shared TigerBeetle requests. The practical limit is the CPU; see Performance.