This document provides an overview of the recommended way to develop and implement logging for components of Kubernetes. For k/kubernetes these conventions should be seen as a strong recommendation, however any Kubernetes or external project is welcome to follow the guidelines below. Most of the ideas can be applied more broadly.

Kubernetes project uses klog for logging. This library was created as a permanent fork of glog, as the original project was no longer developed. With the introduction of Structured Logging and Contextual Logging, Kubernetes is being migrated to use logr as its logging interface. Any changes in klog are done to ensure a smooth transition to logr. Klog is then only used to manage the logging configuration (SetLogger, retrieving the logger) and its logging calls should not be used anymore. ``

For projects that want to integrate with Kubernetes or write the same log format, retrieving a logr.Logger instance from klog will return an implementation that emits log output the same way as the traditional klog logging calls.

Shared libraries, such as client-go, should not log errors themselves but just return error, because client libraries may be used in CLI UIs that wish to control output.

There are two main klog methods for writing logs: klog.InfoS and klog.ErrorS. Both methods are part of a logr compatible interface. For those two functions exists additional flavors, like klog.V(2).InfoS that allows caller to increase log verbosity (making it only available when --v=2 flag is provided to binary) and klog.InfoSDepth with klog.ErorSDepth that give control to change caller information.

Any other non-structured (without S) klog methods like klog.Infof that use old C like string format interface are no longer recommended for use. Those log calls using those methods should be migrated to their structured variants. Please see the Structured Logging Guide for more information on how to migrate such logs.

All structured logging methods accept the log message as a string, along with any number of key/value pairs that you provide via a variadic interface{} argument. As variadic arguments represent key value pairs, they should always be even in count with first element being key of type string and second value of any type matching that key.

Prototype:

func InfoS(message string, keysAndValues ...interface{})
func InfoSDepth(depth int, message string, keysAndValues ...interface{})
func ErrorS(err error, message string, keysAndValues ...interface{})
func ErrorSDepth(depth int, err error, message string, keysAndValues ...interface{})

Example:

klog.InfoS("Received HTTP request", "method", "GET", "URL", "/metrics", "latency", time.Second)

Good practices when writing log messages are:

• Start from a capital letter.
• Do not end the message with a period.
• Use active voice. Use complete sentences when there is an acting subject ("A could not do B") or omit the subject if the subject would be the program itself ("Could not do B").
• Use past tense ("Could not delete B" instead of "Cannot delete B")
• When referring to an object, state what type of object it is. ("Deleted pod" instead of "Deleted")

• klog.ErrorS - Errors should be used to indicate unexpected behaviours in code, like unexpected errors returned by subroutine function calls. Logs generated by ErrorS command may be enhanced with additional debug information (by logging library). Calling ErrorS with nil as error may be acceptable if there is error condition that deserves a stack trace at this origin point.

• klog.InfoS - Structured logs to the INFO log. InfoS should be used for routine logging. It can also be used to log warnings for expected errors (errors that can happen during routine operations). Depending on log severity it's important to pick a proper verbosity level to ensure that consumer is neither under nor overwhelmed by log volume:

  • klog.V(0).InfoS = klog.InfoS - Generally useful for this to always be visible to a cluster operator
    • Programmer errors
    • Logging extra info about a panic
    • CLI argument handling
  • klog.V(1).InfoS - A reasonable default log level if you don't want verbosity.
    • Information about config (listening on X, watching Y)
    • Errors that repeat frequently that relate to conditions that can be corrected (pod detected as unhealthy)
  • klog.V(2).InfoS - Useful steady state information about the service and important log messages that may correlate to significant changes in the system. This is the recommended default log level for most systems.
    • Logging HTTP requests and their exit code
    • System state changing (killing pod)
    • Controller state change events (starting pods)
    • Scheduler log messages
  • klog.V(3).InfoS - Extended information about changes
    • More info about system state changes
  • klog.V(4).InfoS - Debug level verbosity
    • Logging in particularly thorny parts of code where you may want to come back later and check it
  • klog.V(5).InfoS - Trace level verbosity
    • Context to understand the steps leading up to errors and warnings
    • More information for troubleshooting reported issues

As per the comments, the practical default level is V(2). Developers and QE environments may wish to run at V(3) or V(4). If you wish to change the log level, you can pass in -v=X where X is the desired maximum level to log.

Kubernetes supports multiple logging formats to allow users flexibility in picking a format that best matches their needs. As new features and logging formats are proposed, we expect that the Kubernetes community might find itself in disagreement about what formats and features should be officially supported. This section will serve as a developer documentation describing high level goals of each format, giving a holistic vision of logging options and allowing the community to resolve any future disagreement when making changes.

• Default format in Kubernetes.
• Purpose:
  • Maintain backward compatibility with klog format.
  • Human-readable
  • Development
• Features:
  • Minimal support for structured metadata - required to maintain backward compatibility.
  • Marshals object using fmt.Stringer interface allowing to provide a human-readable reference. For example pod="kube-system/kube-dns".
  • Support for multi-line strings with actual line breaks - allows dumping whole yaml objects for debugging.

Logs written using text format will always have standard header, however message format differ based on whether string formatting (Infof, Errorf, ...) or structured logging method was used to call klog (InfoS, ErrorS).

Examples, first written using Infof, in second InfoS was used.

I0528 19:15:22.737538   47512 logtest.go:52] Pod kube-system/kube-dns status was updated to ready
I0528 19:15:22.737538   47512 logtest.go:52] "Pod status was updated" pod="kube-system/kube-dns" status="ready"

Explanation of header:

Lmmdd hh:mm:ss.uuuuuu threadid file:line] msg...

where the fields are defined as follows:
	L                A single character, representing the log level (eg 'I' for INFO)
	mm               The month (zero padded; ie May is '05')
	dd               The day (zero padded)
	hh:mm:ss.uuuuuu  Time in hours, minutes and fractional seconds
	threadid         The space-padded thread ID as returned by GetTID()
	file             The file name
	line             The line number
	msg              The user-supplied message

See more in here

• Requires passing --logging-format=json to enable
• Purpose:
  • Provide structured metadata in efficient way.
  • Optimized for efficient log consumption, processing and querying.
  • Machine-readable
• Feature:
  • Full support for structured metadata, using logr.MarshalLog when available.
  • Multi-line strings are quoted to fit into a single output line.

Example:

{"ts": 1580306777.04728,"v": 4,"msg": "Pod status was updated","pod":{"name": "kube-dns","namespace": "kube-system"},"status": "ready"}

Keys with special meaning:

• ts - timestamp as Unix time (required, float)
• v - verbosity (only for info and not for error messages, int)
• err - error string (optional, string)
• msg - message (required, string)