CHANGES.md
See the [email protected] changelog for details on recent 1.x releases.
Known issues:
(nothing yet)
req.originalUrl for the "url" field per
https://expressjs.com/en/api.html#req.originalUrl. (By @twelve17 and
@kingcody.)os.EOL for newlines in bunyan output, which helps with
some Unix-EOL-naive apps like notepad. (By @bwknight877.)Fix a vulnerability from a crafted argument to 'bunyan -p ARG'
This was reported privately as: https://hackerone.com/reports/902739 bunyan - RCE via insecure command formatting
Previous to this version the 'bunyan' CLI was not escaping a given argument
to the '-p' option before executing ps -A -o pid,command | grep '$ARG'
which could lead to unintended execution.
(This same change is also in [email protected].)
bunyan CLI to not duplicate the "HTTP/1.1 ..." status
line when serializing a "res" field.bunyan CLI:
The bunyan CLI no longer adds a Host header when rendering a client_req
field in a log record. In 1.x it used to do this (using client_req.address
and client_req.port), on the guess that Node.js' http.ClientRequest
handling would add it. However, the guess can be wrong and misleading. It is
better not to guess.bunyan CLI exit handling problems.bunyan errors out if attempting to use -p PID and
file args at the same time.bunyan CLI version
1.0.1 where SIGINT was ignored, such that Ctrl+C could not be used to
terminate bunyan. (By @zbjornson and @davepacheco.)"use strict;") error in some versions of
Safari.Note: Bad release. Use 1.8.8 or later.
Note: Bad release. Use 1.8.7 or later.
safeCycles is too slow when logging large objects.src usage with node v7.log.info(null) crash that resulted from #426 in v1.8.2.log.info({err: err}) results in a "msg" value, just
like log.info(err).Note: Bad release. An addition in this release broke 'rotating-file' usage. Use 1.8.1 or later.
bunyan -p ... (i.e. DTrace integration) on node
4.x and 5.x.[issue #332, pull #355] Ensure stream for type='stream' stream is a writable stream. (By Michael Nisi.)
[issue #344] Fix "rotating-file" Bunyan streams to not miss rotations when configured
for a period greater than approximately 25 days. Before this there was an issue
where periods greater than node.js's maximum setTimeout length would fail to rotate.
(By Martijn Schrage.)
[issue #234, pull #345] Improve bunyan CLI rendering of "res" field
HTTP responses to not show two blank lines for an empty body.
(By Michael Nisi.)
[pull #311, #302, #310] Improve the runtime environment detection to fix running under NW.js. Contributions by Adam Lynch, Jeremy Ruppel, and Aleksey Timchenko.
[pull #318] Add reemitErrorEvents optional boolean for streams added to a
Bunyan logger to control whether an "error" event on the stream will be
re-emitted on the Logger instance.
var log = bunyan.createLogger({
name: 'foo',
streams: [
{
type: 'raw',
stream: new MyCustomStream(),
reemitErrorEvents: true
}
]
});
Before this change, "error" events were re-emitted on file
streams only. The new
behaviour is as follows:
reemitErrorEvents not specified: file streams will re-emit error events
on the Logger instance.reemitErrorEvents: true: error events will be re-emitted on the Logger
for any stream with a .on() function -- which includes file streams,
process.stdout/stderr, and any object that inherits from EventEmitter.reemitErrorEvents: false: error events will not be re-emitted for any
streams.Dev Note: Bunyan Logger objects don't currently have a .close() method
in which registered error event handlers can be unregistered. That means
that a (presumably rare) situation where code adds dozens of Bunyan Logger
streams to, e.g. process.stdout, and with reemitErrorEvents: true, could
result in leaking Logger objects.
Original work for allowing "error" re-emitting on non-file streams is by Marc Udoff in pull #318.
[pull #304, issue #245] Use [Moment.js][momentjs.com] library to handle
bunyan CLI time formatting in some cases, especially to fix display of
local time. It is now required for local time formatting (i.e. bunyan -L
or bunyan --time local). (By David M. Lee.)
[pull #252] Fix errant client_res={} in bunyan CLI rendering, and avoid
extra newlines in client_req rendering in some cases. (By Thomas Heymann.)
[pull #291, issue #303] Fix LOG.child(...) to not override the "hostname"
field of the parent. A use case is when one manually sets "hostname" to
something other than os.hostname(). (By github.com/Cactusbone.)
[issue #325] Allow one to set level: 0 in createLogger to turn on
logging for all levels. (Adapted from #336 by github.com/sometimesalready.)
Add guards (to resolveLevel) so that all "level" values are validated.
Before this, a bogus level like "foo" or -12 or ['some', 'array'] would
silently be accepted -- with undefined results.
Doc updates for #340 and #305.
Update make test to test against node 5, 4, 0.12 and 0.10.
src: true, which was broken in v1.5.0.Note: Bad release. The addition of 'use strict'; broke Bunyan's src: true
feature. Use 1.5.1 instead.
(Bumping minor ver b/c I'm wary of dtrace-provider changes. :)
bunyan defensive on res.header=null.bunyan defensive on res.header as a boolean.bunyan defensive on err.stack not being a string.log.child(...) to work even if the logger is a sub-class
of Bunyan's Logger class.haveNonRawStreams on <logger>.addStream.JSON.stringify throws -- possibly with an enumerable property
getter than throws. By Martin Gausby.Export bunyan.RotatingFileStream which is needed if one wants to
customize it. E.g. see issue #194.
[pull #122] Source Map support for caller line position for the "src" field. This could be interesting for CoffeeScript users of Bunyan. By Manuel Schneider.
[issue #164] Ensure a top-level level given in bunyan.createLogger
is used for given streams. For example, ensure that the following
results in the stream having a DEBUG level:
var log = bunyan.createLogger({
name: 'foo',
level: 'debug',
streams: [
{
path: '/var/tmp/foo.log'
}
]
});
This was broken in the 1.0.1 release. Between that release and 1.3.0 the "/var/tmp/foo.log" stream would be at the INFO level (Bunyan's default level).
[issue #103] bunyan -L (or bunyan --time local) to show local time.
Bunyan log records store time in UTC time. Sometimes it is convenient
to display in local time.
[issue #205] Fix the "The Bunyan CLI crashed!" checking to properly warn of
the common failure case when -c CONDITION is being used.
[issue #210] Export bunyan.nameFromLevel and bunyan.levelFromName. It can
be a pain for custom streams to have to reproduce that.
[issue #100] Gracefully handle the case of an unbound
Logger.{info,debug,...} being used for logging, e.g.:
myEmittingThing.on('data', log.info)
Before this change, bunyan would throw. Now it emits a warning to stderr once, and then silently ignores those log attempts, e.g.:
bunyan usage error: /Users/trentm/tm/node-bunyan/foo.js:12: attempt to log with an unbound log method: `this` is: { _events: { data: [Function] } }
FATAL + 1 to turn logging off. While the standard named log levels are
the golden path, then intention was not to get in the way of using
other level numbers.[issue #157] Restore dtrace-provider as a dependency (in "optionalDependencies").
Dtrace-provider version 0.3.0 add build sugar that should eliminate the
problems from older versions:
The build is not attempted on Linux and Windows. The build spew is
not emitted by default (use V=1 npm install to see it); instead a
short warning is emitted if the build fails.
Also, importantly, the new dtrace-provider fixes working with node v0.11/0.12.
err fields in bunyan CLI output. Before
this change only the fields part of the typical node.js error stack
(err.stack, err.message, err.name) would be emitted, even though
the Bunyan library would typically include err.code and err.signal
in the raw JSON log record.log.info(err) on a logger with no serializers.Note: Bad release. It breaks log.info(err) on a logger with no serializers.
Use version 1.1.2.
log.info(err) to use the log Logger's err
serializer if it has one, instead of always using the core Bunyan err
serializer. (By Mihai Tomescu.)[issues #105, #138, #151] Export <Logger>.addStream(...) and
<Logger>.addSerializers(...) to be able to add them after Logger creation.
Thanks @andreineculau!
[issue #159] Fix bad handling in construtor guard intending to allow
creation without "new": var log = Logger(...). Thanks @rmg!
[issue #156] Smaller install size via .npmignore file.
[issue #126, #161] Ignore SIGINT (Ctrl+C) when processing stdin. ...| bunyan
should expect the preceding process in the pipeline to handle SIGINT. While
it is doing so, bunyan should continue to process any remaining output.
Thanks @timborodin and @jnordberg!
[issue #160] Stop using ANSI 'grey' in bunyan CLI output, because of the
problems that causes with Solarized Dark themes (see
https://github.com/altercation/solarized/issues/220).
[issue #87] Backward incompatible change to -c CODE improving
performance by over 10x (good!), with a backward incompatible change to
semantics (unfortunate), and adding some sugar (good!).
The -c CODE implementation was changed to use a JS function for processing
rather than vm.runInNewContext. The latter was specatularly slow, so
won't be missed. Unfortunately this does mean a few semantic differences in
the CODE, the most noticeable of which is that this is required to
access the object fields:
# Bad. Works with bunyan 0.x but not 1.x.
$ bunyan -c 'pid === 123' foo.log
...
# Good. Works with all versions of bunyan
$ bunyan -c 'this.pid === 123' foo.log
...
The old behaviour of -c can be restored with the BUNYAN_EXEC=vm
environment variable:
$ BUNYAN_EXEC=vm bunyan -c 'pid === 123' foo.log
...
Some sugar was also added: the TRACE, DEBUG, ... constants are defined, so one can:
$ bunyan -c 'this.level >= ERROR && this.component === "http"' foo.log
...
And example of the speed improvement on a 10 MiB log example:
$ time BUNYAN_EXEC=vm bunyan -c 'this.level === ERROR' big.log | cat >slow
real 0m6.349s
user 0m6.292s
sys 0m0.110s
$ time bunyan -c 'this.level === ERROR' big.log | cat >fast
real 0m0.333s
user 0m0.303s
sys 0m0.028s
The change was courtesy Patrick Mooney (https://github.com/pfmooney). Thanks!
Add bunyan -0 ... shortcut for bunyan -o bunyan ....
[issue #135] Backward incompatible. Drop dtrace-provider even from
optionalDependencies. Dtrace-provider has proven a consistent barrier to
installing bunyan, because it is a binary dep. Even as an optional dep it
still caused confusion and install noise.
Users of Bunyan on dtrace-y platforms (SmartOS, Mac, Illumos, Solaris) will
need to manually npm install dtrace-provider themselves to get Bunyan's
dtrace support
to work. If not installed, bunyan should stub it out properly.
[issue #139] Fix bunyan crash on a log record with res.header that is an
object. A side effect of this improvement is that a record with res.statusCode
but no header info will render a response block, for example:
[2012-08-08T10:25:47.637Z] INFO: my-service/12859 on my-host: some message (...)
...
--
HTTP/1.1 200 OK
--
...
[pull #42] Fix bunyan crash on a log record with req.headers that is a string
(by https://github.com/aexmachina).
Drop node 0.6 support. I can't effectively npm install with a node 0.6
anymore.
[issue #85] Ensure logging a non-object/non-string doesn't throw (by https://github.com/mhart). This changes fixes:
log.info(<bool>) # TypeError: Object.keys called on non-object
log.info(<function>) # "msg":"" (instead of wanted "msg":"[Function]")
log.info(<array>) # "msg":"" (instead of wanted "msg":util.format(<array>))
Note: Bad release. The published package in the npm registry got corrupted. Use 0.22.3 or later.
[issue #131] Allow log.info(<number>) and, most importantly, don't crash on that.
Update 'mv' optional dep to latest.
[issue #111] Fix a crash when attempting to use bunyan -p on a platform without
dtrace.
[issue #101] Fix a crash in bunyan rendering a record with unexpected "res.headers".
log.reopenFileStreams() convenience method to be used with external log
rotation.bunyan to default to paging (with less) by default in node 0.10.0.
The intention has always been to default to paging for node >=0.8.bunyan -p '*' breakage in version 0.21.2.Note: Bad release. The switchrate change below broke bunyan -p '*' usage
(see issue #90). Use 0.21.3 or later.
[issue #88] Should be able to efficiently combine "-l" with "-p *".
Avoid DTrace buffer filling up, e.g. like this:
$ bunyan -p 42241 > /tmp/all.log
dtrace: error on enabled probe ID 3 (ID 75795: bunyan42241:mod-87ea640:log-trace:log-trace): out of scratch space in action #1 at DIF offset 12
dtrace: error on enabled probe ID 3 (ID 75795: bunyan42241:mod-87ea640:log-trace:log-trace): out of scratch space in action #1 at DIF offset 12
dtrace: 138 drops on CPU 4
...
From Bryan: "the DTrace buffer is filling up because the string size is so large... by increasing the switchrate, you're increasing the rate at which that buffer is emptied."
period (by github.com/ricardograca).[Slight backward incompatibility] Fix serializer bug introduced in 0.18.3 (see below) to only apply serializers to log records when appropriate.
This also makes a semantic change to custom serializers. Before this change
a serializer function was called for a log record key when that value was
truth-y. The semantic change is to call the serializer function as long
as the value is not undefined. That means that a serializer function
should handle falsey values such as false and null.
Update to latest 'mv' dep (required for rotating-file support) to support node v0.10.0.
WARNING: This release includes a bug introduced in bunyan 0.18.3 (see below). Please upgrade to bunyan 0.20.0.
[Slight backward incompatibility] Change the default error serialization
(a.k.a. bunyan.stdSerializers.err) to not serialize all additional
attributes of the given error object. This is an open door to unsafe logging
and logging should always be safe. With this change, error serialization
will log these attributes: message, name, stack, code, signal. The latter
two are added because some core node APIs include those fields (e.g.
child_process.exec).
Concrete examples where this has hurt have been the "domain" change
necessitating 0.18.3 and a case where
node-restify uses an error object
as the response object. When logging the err and res in the same log
statement (common for restify audit logging), the res.body would be JSON
stringified as '[Circular]' as it had already been emitted for the err key.
This results in a WTF with the bunyan CLI because the err.body is not
rendered.
If you need the old behaviour back you will need to do this:
var bunyan = require('bunyan');
var errSkips = {
// Skip domain keys. `domain` especially can have huge objects that can
// OOM your app when trying to JSON.stringify.
domain: true,
domain_emitter: true,
domain_bound: true,
domain_thrown: true
};
bunyan.stdSerializers.err = function err(err) {
if (!err || !err.stack)
return err;
var obj = {
message: err.message,
name: err.name,
stack: getFullErrorStack(err)
}
Object.keys(err).forEach(function (k) {
if (err[k] !== undefined && !errSkips[k]) {
obj[k] = err[k];
}
});
return obj;
};
"long" and "bunyan" output formats for the CLI. bunyan -o long is the default
format, the same as before, just called "long" now instead of the cheesy "paul"
name. The "bunyan" output format is the same as "json-0", just with a more
convenient name.
WARNING: This release introduced a bug such that all serializers are
applied to all log records even if the log record did not contain the key
for that serializer. If a logger serializer function does not handle
being given undefined, then you'll get warnings like this on stderr:
bunyan: ERROR: This should never happen. This is a bug in <https://github.com/trentm/node-bunyan> or in this application. Exception from "foo" Logger serializer: Error: ...
at Object.bunyan.createLogger.serializers.foo (.../myapp.js:20:15)
at Logger._applySerializers (.../lib/bunyan.js:644:46)
at Array.forEach (native)
at Logger._applySerializers (.../lib/bunyan.js:640:33)
...
and the following junk in written log records:
"foo":"(Error in Bunyan log "foo" serializer broke field. See stderr for details.)"
Please upgrade to bunyan 0.20.0.
Change the bunyan.stdSerializers.err serializer for errors to exclude
the "domain*" keys.
err.domain will include its assigned members which can arbitrarily large
objects that are not intended for logging.
Make the "dtrace-provider" dependency optional. I hate to do this, but installing bunyan on Windows is made very difficult with this as a required dep. Even though "dtrace-provider" stubs out for non-dtrace-y platforms, without a compiler and Python around, node-gyp just falls over.
bunyan CLI to not automatically page (i.e. pipe to less)
if stdin isn't a TTY, or if following dtrace probe output (via -p PID),
or if not given log file arguments.Automatic paging support in the bunyan CLI (similar to git log et al).
IOW, bunyan will open your pager (by default less) and pipe rendered
log output through it. A main benefit of this is getting colored logs with
a pager without the pain. Before you had to explicit use --color to tell
bunyan to color output when the output was not a TTY:
bunyan foo.log --color | less -R # before
bunyan foo.log # now
Disable with the --no-pager option or the BUNYAN_NO_PAGER=1 environment
variable.
Limitations: Only supported for node >=0.8. Windows is not supported (at least not yet).
Switch test suite to nodeunit (still using a node-tap'ish API via a helper).
[issue #33] Log rotation support:
var bunyan = require('bunyan');
var log = bunyan.createLogger({
name: 'myapp',
streams: [{
type: 'rotating-file',
path: '/var/log/myapp.log',
count: 7,
period: 'daily'
}]
});
Tweak to CLI default pretty output: don't special case "latency" field. The special casing was perhaps nice, but less self-explanatory. Before:
[2012-12-27T21:17:38.218Z] INFO: audit/45769 on myserver: handled: 200 (15ms, audit=true, bar=baz)
GET /foo
...
After:
[2012-12-27T21:17:38.218Z] INFO: audit/45769 on myserver: handled: 200 (audit=true, bar=baz, latency=15)
GET /foo
...
Exit CLI on EPIPE, otherwise we sit there useless processing a huge log
file with, e.g. bunyan huge.log | head.
Guards on -c CONDITION usage to attempt to be more user friendly.
Bogus JS code will result in this:
$ bunyan portal.log -c 'this.req.username==boo@foo'
bunyan: error: illegal CONDITION code: SyntaxError: Unexpected token ILLEGAL
CONDITION script:
Object.prototype.TRACE = 10;
Object.prototype.DEBUG = 20;
Object.prototype.INFO = 30;
Object.prototype.WARN = 40;
Object.prototype.ERROR = 50;
Object.prototype.FATAL = 60;
this.req.username==boo@foo
Error:
SyntaxError: Unexpected token ILLEGAL
at new Script (vm.js:32:12)
at Function.Script.createScript (vm.js:48:10)
at parseArgv (/Users/trentm/tm/node-bunyan-0.x/bin/bunyan:465:27)
at main (/Users/trentm/tm/node-bunyan-0.x/bin/bunyan:1252:16)
at Object.<anonymous> (/Users/trentm/tm/node-bunyan-0.x/bin/bunyan:1330:3)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:492:10)
And all CONDITION scripts will be run against a minimal valid Bunyan log record to ensure they properly guard against undefined values (at least as much as can reasonably be checked). For example:
$ bunyan portal.log -c 'this.req.username=="bob"'
bunyan: error: CONDITION code cannot safely filter a minimal Bunyan log record
CONDITION script:
Object.prototype.TRACE = 10;
Object.prototype.DEBUG = 20;
Object.prototype.INFO = 30;
Object.prototype.WARN = 40;
Object.prototype.ERROR = 50;
Object.prototype.FATAL = 60;
this.req.username=="bob"
Minimal Bunyan log record:
{
"v": 0,
"level": 30,
"name": "name",
"hostname": "hostname",
"pid": 123,
"time": 1355514346206,
"msg": "msg"
}
Filter error:
TypeError: Cannot read property 'username' of undefined
at bunyan-condition-0:7:9
at Script.Object.keys.forEach.(anonymous function) [as runInNewContext] (vm.js:41:22)
at parseArgv (/Users/trentm/tm/node-bunyan-0.x/bin/bunyan:477:18)
at main (/Users/trentm/tm/node-bunyan-0.x/bin/bunyan:1252:16)
at Object.<anonymous> (/Users/trentm/tm/node-bunyan-0.x/bin/bunyan:1330:3)
at Module._compile (module.js:449:26)
at Object.Module._extensions..js (module.js:467:10)
at Module.load (module.js:356:32)
at Function.Module._load (module.js:312:12)
at Module.runMain (module.js:492:10)
A proper way to do that condition would be:
$ bunyan portal.log -c 'this.req && this.req.username=="bob"'
[issue #56] Support bunyan -p NAME to dtrace all PIDs matching 'NAME' in
their command and args (using ps -A -o pid,command | grep NAME or, on SunOS
pgrep -lf NAME). E.g.:
bunyan -p myappname
This is useful for usage of node's cluster module where you'll have multiple worker processes.
bunyan -p '*' to capture bunyan dtrace probes from all processes.BUNYAN_NO_COLOR environment variable to
turn off all output coloring. This is still overridden by the --color
and --no-color options.bunyan -p PID) gets through. There had been a race
between exiting bunyan and the flushing of the dtrace process' stderr.bunyan -p PID)
gets through. The pipe usage wasn't working on SmartOS. This is important
to show the user if they need to 'sudo'.bunyan -p PID) is
terminated on signal termination of the bunyan CLI (at least for SIGINT,
SIGQUIT, SIGTERM, SIGHUP).Add bunyan -p PID support. This is a convenience wrapper that effectively
calls:
dtrace -x strsize=4k -qn 'bunyan$PID:::log-*{printf("%s", copyinstr(arg0))}' | bunyan
issue #48: Dtrace support! The elevator pitch is you can watch all logging from all Bunyan-using process with something like this:
dtrace -x strsize=4k -qn 'bunyan*:::log-*{printf("%d: %s: %s", pid, probefunc, copyinstr(arg0))}'
And this can include log levels below what the service is actually configured to log. E.g. if the service is only logging at INFO level and you need to see DEBUG log messages, with this you can. Obviously this only works on dtrace-y platforms: Illumos derivatives of SunOS (e.g. SmartOS, OmniOS), Mac, FreeBSD.
Or get the bunyan CLI to render logs nicely:
dtrace -x strsize=4k -qn 'bunyan*:::log-*{printf("%s", copyinstr(arg0))}' | bunyan
See https://github.com/trentm/node-bunyan#dtrace-support for details. By Bryan Cantrill.
Export bunyan.safeCycles(). This may be useful for custom type == "raw"
streams that may do JSON stringification of log records themselves. Usage:
var str = JSON.stringify(rec, bunyan.safeCycles());
[issue #49] Allow a log.child() to specify the level of inherited streams.
For example:
# Before
var childLog = log.child({...});
childLog.level('debug');
# After
var childLog = log.child({..., level: 'debug'});
Improve the Bunyan CLI crash message to make it easier to provide relevant details in a bug report.
Fix a bug in the long-stack-trace error serialization added in 0.14.4. The symptom:
[email protected]: .../node_modules/bunyan/lib/bunyan.js:1002
var ret = ex.stack || ex.toString();
^
TypeError: Cannot read property 'stack' of undefined
at getFullErrorStack (.../node_modules/bunyan/lib/bunyan.js:1002:15)
...
Bad release. Use 0.14.5 instead.
Improve error serialization to walk the chain of .cause() errors
from the likes of WError or VError error classes from
verror and
restify v2.0. Example:
[2012-10-11T00:30:21.871Z] ERROR: imgapi/99612 on 0525989e-2086-4270-b960-41dd661ebd7d: my-message
ValidationFailedError: my-message; caused by TypeError: cause-error-message
at Server.apiPing (/opt/smartdc/imgapi/lib/app.js:45:23)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server.setupReq (/opt/smartdc/imgapi/lib/app.js:178:9)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server.parseBody (/opt/smartdc/imgapi/node_modules/restify/lib/plugins/body_parser.js:15:33)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server.parseQueryString (/opt/smartdc/imgapi/node_modules/restify/lib/plugins/query.js:40:25)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server._run (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:579:17)
at Server._handle.log.trace.req (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:480:38)
Caused by: TypeError: cause-error-message
at Server.apiPing (/opt/smartdc/imgapi/lib/app.js:40:25)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server.setupReq (/opt/smartdc/imgapi/lib/app.js:178:9)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server.parseBody (/opt/smartdc/imgapi/node_modules/restify/lib/plugins/body_parser.js:15:33)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server.parseQueryString (/opt/smartdc/imgapi/node_modules/restify/lib/plugins/query.js:40:25)
at next (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:550:50)
at Server._run (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:579:17)
at Server._handle.log.trace.req (/opt/smartdc/imgapi/node_modules/restify/lib/server.js:480:38)
bunyan CLI output of a res.body that is an
object instead of a string. See issue#38 for the same with req.body.JSON.stringifying of emitted log records to avoid blowing
up on circular objects (by Isaac Schlueter).client_req handling in the default output
of the bunyan CLI.bunyan CLI output of a req.body that is an
object instead of a string.Export bunyan.resolveLevel(NAME-OR-NUM) to resolve a level name or number
to its log level number value:
> bunyan.resolveLevel('INFO')
30
> bunyan.resolveLevel('debug')
20
A side-effect of this change is that the uppercase level name is now allowed in the logger constructor.
log.info(BUFFER), where BUFFER is
a node.js Buffer object, doesn't blow up.req.body, res.body and other request/response fields
are emitted by the bunyan CLI (mostly by Rob Gulewich).[issue #31] Re-instate defines for the (uppercase) log level names (TRACE,
DEBUG, etc.) in bunyan -c "..." filtering condition code. E.g.:
$ ... | bunyan -c 'level >= ERROR'
[pull #32] bunyan -o short for more concise output (by Dave Pacheco). E.g.:
22:56:52.856Z INFO myservice: My message
instead of:
[2012-02-08T22:56:52.856Z] INFO: myservice/123 on example.com: My message
bunyan CLI to suppress all but legal Bunyan JSON
log lines. By default non-JSON, and non-Bunyan lines are passed through.bunyan CLI.bunyan -c "..."
filtering code. This was added in v0.11.1, but has a significant adverse
affect.Bad release. The TRACE et al names are bleeding into the log records when using '-c'.
Add defines for the (uppercase) log level names (TRACE, DEBUG, etc.) in
bunyan -c "..." filtering condition code. E.g.:
$ ... | bunyan -c 'level >= ERROR'
[pull #29] Add -l/--level for level filtering, and -c/--condition for arbitrary conditional filtering (by github.com/isaacs):
$ ... | bunyan -l error # filter out log records below error
$ ... | bunyan -l 50 # numeric value works too
$ ... | bunyan -c 'level===50' # equiv with -c filtering
$ ... | bunyan -c 'pid===123' # filter on any field
$ ... | bunyan -c 'pid===123' -c '_audit' # multiple filters
[pull #24] Support for gzip'ed log files in the bunyan CLI (by github.com/mhart):
$ bunyan foo.log.gz
...
[pull #16] Bullet proof the bunyan.stdSerializers (by github.com/rlidwka).
[pull #15] The bunyan CLI will now chronologically merge multiple log
streams when it is given multiple file arguments. (by github.com/davepacheco)
$ bunyan foo.log bar.log
... merged log records ...
[pull #15] A new bunyan.RingBuffer stream class that is useful for
keeping the last N log messages in memory. This can be a fast way to keep
recent, and thus hopefully relevant, log messages. (by @dapsays,
github.com/davepacheco)
Potential uses: Live debugging if a running process could inspect those
messages. One could dump recent log messages at a finer log level than is
typically logged on
uncaughtException.
var ringbuffer = new bunyan.RingBuffer({ limit: 100 });
var log = new bunyan({
name: 'foo',
streams: [{
type: 'raw',
stream: ringbuffer,
level: 'debug'
}]
});
log.info('hello world');
console.log(ringbuffer.records);
Add support for "raw" streams. This is a logging stream that is given raw log record objects instead of a JSON-stringified string.
function Collector() {
this.records = [];
}
Collector.prototype.write = function (rec) {
this.records.push(rec);
}
var log = new Logger({
name: 'mylog',
streams: [{
type: 'raw',
stream: new Collector()
}]
});
See "examples/raw-stream.js". I expect raw streams to be useful for piping Bunyan logging to separate services (e.g. http://www.loggly.com/, https://github.com/etsy/statsd) or to separate in-process handling.
Add test/corpus/*.log files (accidentally excluded) so the test suite actually works(!).
[pull #21] Bunyan loggers now re-emit fs.createWriteStream error events.
By github.com/EvanOxfeld. See "examples/handle-fs-error.js" and
"test/error-event.js" for details.
var log = new Logger({name: 'mylog', streams: [{path: FILENAME}]});
log.on('error', function (err, stream) {
// Handle error writing to or creating FILENAME.
});
jsstyle'ing (via make check)
bunyan.createLogger(OPTIONS) form, as is more typical in
node.js APIs. This'll eventually become the preferred form.Change bunyan CLI default output to color "src" info red. Before the "src"
information was uncolored. The "src" info is the filename, line number and
function name resulting from using src: true in Logger creation. I.e.,
the (/Users/trentm/tm/node-bunyan/examples/hi.js:10) in:
[2012-04-10T22:28:58.237Z] INFO: myapp/39339 on banana.local (/Users/trentm/tm/node-bunyan/examples/hi.js:10): hi
Tweak bunyan CLI default output to still show an "err" field if it doesn't
have a "stack" attribute.
log.child({...}, true); where the added child fields would
be added to the parent's fields. This bug only existed for the "fast child"
path (that second true argument). A side-effect of fixing this is that
the "fast child" path is only 5 times as fast as the regular log.child,
instead of 10 times faster.bunyan CLI taking log file path args, bunyan foo.log,
in addition to the usual cat foo.log | bunyan.bunyan CLI.
Before it could blow up processing log records missing some expected
fields.bunyan CLI tool (for the default output mode/style).
Also add the '--color' option to force coloring if the output stream is not
a TTY, e.g. cat my.log | bunyan --color | less -R. Use --no-color to
disable coloring, e.g. if your terminal doesn't support ANSI codes.log.info() -> boolean to work properly. Previous all were
returning false. Ditto all trace/debug/.../fatal methods.Allow an optional msg and arguments to the log.info(<Error> err) logging
form. For example, before:
log.debug(my_error_instance) // good
log.debug(my_error_instance, "boom!") // wasn't allowed
Now the latter is allowed if you want to expliciting set the log msg. Of course
this applies to all the log.{trace|debug|info...}() methods.
bunyan cli output: clarify extra fields with quoting if empty or have
spaces. E.g. 'cmd' and 'stderr' in the following:
[2012-02-12T00:30:43.736Z] INFO: mo-docs/43194 on banana.local: buildDocs results (req_id=185edca2-2886-43dc-911c-fe41c09ec0f5, route=PutDocset, error=null, stderr="", cmd="make docs")
bunyan CLI
default output.bunyan CLI default output to include all log record keys. Unknown keys
are either included in the first line parenthetical (if short) or in the indented
subsequent block (if long or multiline).new Logger(...) and log.child(...)
options.[issue #2] Add guard on JSON.stringifying of log records before emission.
This will prevent log.info et al throwing on record fields that cannot be
represented as JSON. An error will be printed on stderr and a clipped log
record emitted with a 'bunyanMsg' key including error details. E.g.:
bunyan: ERROR: could not stringify log record from /Users/trentm/tm/node-bunyan/examples/unstringifyable.js:12: TypeError: Converting circular structure to JSON
{
"name": "foo",
"hostname": "banana.local",
"bunyanMsg": "bunyan: ERROR: could not stringify log record from /Users/trentm/tm/node-bunyan/examples/unstringifyable.js:12: TypeError: Converting circular structure to JSON",
...
Some timing shows this does effect log speed:
$ node tools/timeguard.js # before
Time try/catch-guard on JSON.stringify:
- log.info: 0.07365ms per iteration
$ node tools/timeguard.js # after
Time try/catch-guard on JSON.stringify:
- log.info: 0.07368ms per iteration
Use 10/20/... instead of 1/2/... for level constant values. Ostensibly this allows for intermediary levels from the defined "trace/debug/..." set. However, that is discouraged. I'd need a strong user argument to add support for easily using alternative levels. Consider using a separate JSON field instead.
s/service/name/ for Logger name field. "service" is unnecessarily tied to usage for a service. No need to differ from log4j Logger "name".
Add log.level(...) and log.levels(...) API for changing logger stream
levels.
Add TRACE|DEBUG|INFO|WARN|ERROR|FATAL level constants to exports.
Add log.info(err) special case for logging an Error instance. For
example log.info(new TypeError("boom") will produce:
...
"err": {
"message": "boom",
"name": "TypeError",
"stack": "TypeError: boom\n at Object.<anonymous> ..."
},
"msg": "boom",
...
Add new Logger({src: true}) config option to have a 'src' attribute be
automatically added to log records with the log call source info. Example:
"src": {
"file": "/Users/trentm/tm/node-bunyan/examples/src.js",
"line": 20,
"func": "Wuzzle.woos"
},
log.child(options[, simple]) Added simple boolean arg. Set true to
assert that options only add fields (no config changes). Results in a 10x
speed increase in child creation. See "tools/timechild.js". On my Mac,
"fast child" creation takes about 0.001ms. IOW, if your app is dishing
10,000 req/s, then creating a log child for each request will take
about 1% of the request time.log.clone -> log.child to better reflect the relationship: streams and
serializers are inherited. Streams can't be removed as part of the child
creation. The child doesn't own the parent's streams (so can't close them).log.child usage
is fast. TODO: measure that.Logger.stdSerializers.err serializer which is necessary to get good
Error object logging with node 0.6 (where core Error object properties
are non-enumerable).LOG_VERSION to exports.First release.