Developers · Best Practices

Back-End Performance

Last updated 1 November 2026

Find what makes a page slow: log from your code, read the log, and time each part with time_diff.

Back-end performance is how long your instance takes to build a page before it sends it. Most of that time is spent in Liquid and in GraphQL queries. This section shows you what your code is doing and how long each part takes. Then you can fix the slow part instead of guessing.

The tools

Tool	What it gives you
The `log` tag	Writes any value to the instance log, with a type you choose, from pages, layouts, partials and background jobs
`insites-cli logs <environment>`	Follows the instance log in your terminal while you load the page
The `admin_logs` query	Reads the same log entries from GraphQL, 1 to 500 at a time
The `to_time` and `time_diff` filters	Measure the time between two points in your code

Example: time one query

Take the time before the code you want to measure, then the difference after it. time_diff returns milliseconds unless you give a unit:

{%- assign started = 'now' | to_time -%}
{%- graphql result -%}
  query { records(per_page: 50, filter: { table: { value: "modules/website/developers_guide" } }) { total_entries } }
{%- endgraphql -%}
{%- assign ms = started | time_diff: 'now' -%}
pages={{ result.records.total_entries }} ms={{ ms }}

insites-cli exec liquid docs57 -f timing.liquid
pages=213 ms=18.085

This ran on the docs.insites.io instance on 1 November 2026. In a real page, write the result to the log with the log tag instead of printing it, then read it with insites-cli logs.

Where to look first

GraphQL calls inside a loop. Each one is a round trip; fetch the related data in one query instead.
Large per_page values when you only show a few rows.
Reports and exports that read a lot of data. On records and users, options: { read_replica: true } runs the read on a replica when the environment has one. Do not use it straight after a write, because a replica can lag.

Common mistakes

Dividing a bare time_diff by 3600 to get hours. The result is in milliseconds; pass 'h' instead.
Logging a constant's value while you debug. The log is readable by anyone with access to it.
Measuring once. Load the page several times and compare.