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.
The primary 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 that 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 ffx log is a good default to see all logs. Under the hood, this
command runs the log_listener program, printing every message from the system
log. This includes those forwarded from the klog.
log_listener receives logs through the fuchsia.logger.Log and
fuchsia.logger.LogListenerSafe protocols.
Additionally, some logs from syslog are printed to the serial console. By default, this includes the
driver and driver_manager logs.
log_listener emits lines in this format by default:
[seconds][pid][tid][tags] LEVEL: message
The timestamp is from the monotonic clock by default, and it is formatted with microsecond granularity.
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][1902][1904][my-component.cmx] WARN: something happened
log_listener has --hide_metadata and --pretty flags that reduces the printed metadata,
and color codes log lines by severity, respectively. With these flags, some metadata is hidden
(PID, TID, etc.) while others are trimmed down (timestamp, severity).
For example, if the message "something happened" is printed at WARN level by my-component.cmx at time=278.14, the pretty output will look like:
[278.14][my-component.cmx][W] something happened
With a running device available, run ffx log --help to see the options for modifying the output format.
Under the hood, fx test calls run-test-component, which collects isolated stdout, stderr, and
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 ffx log.
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
The fx pretty_serial command can be used to reduce the metadata printed by klog and color code
log lines by severity. With this command, some metadata is hidden (PID, TID, filenames, etc.)
while others are trimmed down (timestamp, severity).
Serial output should be piped in from the emulator or from other sources:
ffx emu start --console | fx symbolize
For example, if the message "something happened" is printed to klog at WARN level by my-component.cmx at time=278.14, the pretty output will look like:
[278.14][my-component.cmx][W] something happened
For example, if the message "something happened" is printed to klog by an unknown component with unknown severity at time=278.14, the pretty output will look like:
[278.14] something happened
When running tests, a Swarming bot invokes botanist, which collects several output streams to be
presented in the web UI. The stdout & 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.
Botanist runs log_listener on the target device and saves that output to syslog.txt. This is
comparable to running ffx 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:
- botanist's log messages
- kernel log from netsvc (equivalent to
fx klog) - testrunner's log messages
stdoutandstderrof the tests run by testrunner
This aggregate log is run through the equivalent of fx symbolize before upload.