jq

. is the identity filter: output equals input. jq re-indents it with two spaces and one key per line. It also validates: malformed JSON is an error with a non-zero exit status.

.version selects the key from the top-level object. The result is a JSON string, so it keeps its double quotes.

Paths chain with dots: .database.host. -r prints strings raw, without JSON quotes, so the value can go straight into a shell variable: host=$(jq -r .database.host config.json).

A key that does not exist gives null, not an error. The alternative operator // substitutes a default when the left side is null or false.

servers.json is an array. .[] emits each element separately, and .name runs on each one, giving five outputs. With -r they print one per line, ready for while read.

select(cond) passes an element through only when cond is true, otherwise drops it. Only db01 and build01 have eight or more CPUs.

Conditions combine with and, or and not. index("us") on an array returns the position of that element or null, and null counts as false. {name, tags} is shorthand for {name: .name, tags: .tags}. -c prints each object on one line.

Object construction picks and renames fields. "\(...)" is string interpolation. The prod value is computed: contains(["prod"]) tests whether the tags array includes that string.

map(.cpus) turns the array of servers into an array of numbers. The three filters after it each receive that array: add sums it to 34, max finds 16, and dividing the sum by length gives the average.

On an object, length counts its keys and keys lists them sorted alphabetically (use keys_unsorted to keep file order). On an array, length counts elements.

--arg role "$role" defines $role inside the filter as a JSON string. The filter itself stays in single quotes, so the shell never touches it. Use --argjson when the value must be a number or other JSON, for example --argjson min 8.

Build an array per element, then @tsv formats it as one tab-separated line (@csv gives quoted CSV). With -r the result is plain text that column -t, sort or awk can use.

group_by(.role) sorts by role and returns an array of arrays, one per distinct role. map turns each group into an object with the role and the group's size.

= sets a path and returns the whole modified document. jq never edits in place: write to a new file and move it over the old one. Redirecting straight back into data/config.json would truncate it before jq reads it.

The trailing comma is not valid JSON. jq reports where parsing failed and exits with status 5, so set -e scripts and if jq ... tests notice the bad input.