In many projects we use JSON as log format since it inheritly bears structure and thus is easy to process by log-aggregators (i.e. AWS CloudWatch autom. discovers fields from json just like fluentd).
And still, although those aggregators are live safers when skimming through tons of logs I’m often faster investigating an isolated issue using the good old log-tailing. While this was already a doubtful pleasure in the past it became a real pain now with JSON as output format.
In order to ease that pain and bring back the pleasure of bygone days I combine jq (a CLI json-processor I already wrote about here) with stern (a multi pod log tailing tool for kubernetes).
# ┌─> Tells `stern` to output the logs as-is (without prepending the
# │ pod-name - which would only irritate `jq`)
# │ ┌─> Common sub-string of pod-names to be tailed (be just as
# │ │ precise as you have to be - i.e. 'foo' will tail all pods
# │ │ that contain that string in their name)
# │ │ ┌─> start with the last 10 lines
# │ │ │ ┌─> some logs are still not JSON ... find
# │ │ │ │ a common marker that all JSON-logs have
# │ │ │ │ in common
# ┌─┴──┐ ┌─┴─────┐ ┌─┴─────┐ ┌─┴───────────┐
$> stern -o raw <POD-NAME> --tail 10 -i 'LogMessage' | \
jq '. | (.ThreadID +" | "+ .LogMessage)'
# └──┬──────────────────────────┘
# └─> we extract only the 'TreadID' and 'LogMessage' props from
# the original log, concatenate 'em intermediary structure
stern
will tail the logs of all matching pods to stdout. That stream of logs will then be piped into jq
to select and reformat the required information.
Using the flexibility of jq
you can further add select-filters like:
$> stern -o raw <POD-NAME> --tail 10 -i 'LogMessage' | \
jq 'select(.LogMessage|test("Exiting")) | .LogMessage'
# └─┬───────────────────────────────┘
# └─> regex-test to filter only those entries that
# contain the word 'Exiting' somewhere in the
# 'LogMessage's value
$> stern -o raw <POD-NAME> --tail 10 -i 'LogMessage' | \
jq 'select(.ThreadID=="84") | .LogMessage'
# └─┬───────────────────┘
# └─> outputs only those entries that have a 'ThreadID'
# of '84' (this is an exact match)
These examples would directly apply to Payara’s json log format but can easily be adjusted to match different formats.