Logs are primarily consumed in either an interactive (“online”) context with a live device, or in an “offline” context with logs collected from past execution of a device.
All logs have a timestamp attached which is read from the monotonic clock when recording the message. There are many ways that a
LogSink can receive messages in a different order than indicated by their timestamps.
fuchsia.logger.Log implementation sorts messages it is sending via the
LogMany method on
fuchsia.logger.LogListenerSafe. This method is called whenever sending already-collected messages to a newly-connected listener. However, when messages first arrive with out-of-order timestamps, any active listeners will receive them in an arbitrary ordering. Tools which display logs accumulated from successive calls to
fuchsia.logger/LogListenerSafe.Log should alert their users when messages are received without a strictly linear ordering in timestamps.
Because there are two buffers which store logs, there are two main ways to view them when you have a live device. For more information about where logs are stored on-device, see Concepts: Storage.
During development, running
fx log is a good default to see all logs. This script connects to the target device with SSH and runs the
log_listener program, printing every message from the system log. This includes those forwarded from the klog.
log_listener emits lines in this format by default:
[seconds.nanos][pid][tid][tags] LEVEL: message
The timestamp is from the monotonic clock by default.
If the message “something happened” is written at WARN level by my-component.cmx from process=1902 and thread=1904 at time=278.14, the default output would be:
[278.14][my-component.cmx] WARN: something happened
With a running device available, run
fx log -h to see the options for modifying the output format.
Under the hood,
fx test calls
run-test-component which collects isolated
LogSink connections from test components, printing the output inline and preventing them showing up in the global log buffers.
For tests that are not yet components no interception of logs is performed.
The klog is printed over the kernel console and serial.
It‘s also forwarded over UDP by netsvc, which is what’s printed when you run
fx klog. Running
fx klog in a background terminal can be a good way to capture logs if your SSH session fails, or as a backup if there are other issues with running
If neither of the above are options, you can also use
dlog from a device shell directly to dump the kernel debug logs.
The kernel log's dumper emits lines in the format:
[timestamp] pid:tid> message
The timestamp is from the monotonic clock. It is formatted with 5 digits (leading zeroes) for seconds and three digits for milliseconds (trailing zeroes).
Process and thread koids are written with 5 digits each (leading zeroes).
If the message “something happened” is written from process=1902 and thread=1904 at time=278.14, the resulting output would be:
[00278.140] 01902:01904> something happened
When running tests, a Swarming bot invokes botanist which collects several output streams to be presented in the web UI. The
stderr of botanist are what's presented in the “swarming task UI”.
For individual test executables botanist invokes testrunner and collects that output separately. It is this output that can be seen after a failing test, with a link named
stdio. Most tests that testrunner invokes run
run-test-component via SSH to the target device. This collects the stdout, stderr, and logs from the test environment and prints them inline.
log_listener on the target device and saves that output to syslog.txt. This is comparable to running
fx log at a development machine.
This log includes the stdout and stderr of the command run by the Swarming task. Normally this includes the following notable items, all interleaved:
stderrof the tests run by testrunner
This aggregate log is run through the equivalent of
fx symbolize before upload.