Chronicles

Chronicles is a library for structured logging. It adheres to the philosophy that log files shouldn't be based on formatted text strings, but rather on well-defined event records with arbitrary properties that are easy to read for both humans and machines. Let's illustrate this with an example:

import net, chronicles

socket.accept(...)
...
debug "Client PSK", psk = client.getPskIdentity
info "New incoming connection", remoteAddr = ip, remotePort = port

1
2
3
4
5
6
7

Here, debug and info are logging statements, corresponding to different severity levels. You can think of their first argument as the name of a particular event that happened during the execution of the program, while the rest of the arguments are the properties of this event.

From these logging statements, Chronicles can be configured to produce log output in various structured formats. The default format is called textlines and it looks like this:

textblocks format example

Alternatively, you can use a multi-line format called textblocks:

textblocks format example

While these human-readable formats provide a more traditional and familiar experience of using a logging library, the true power of Chronicles is unlocked only after switching to the JSON format. Then, the same log output will look like this:

json format example

At first, switching to JSON may look like a daunting proposition, but Chronicles provides a customized log tailing program called chronicles-tail which is able to transform the JSON stream back into the familiar human-readable form, while also providing additional advanced features such as on on-the-fly filtering, sound alerts and more.

The main advantage of using JSON logging is that this facilitates the storage of the log records in specialized databases which are usually able to provide search and filtering capabilities and allow you to compute various aggregated metrics and time-series data from the accumulated logs.

Typical log storage choices for the above are open-source search engines such as ElasticSearch or specialized providers such as Loggly.

Logging Scopes

In the introduction, we saw debug and info as examples for logging statements. Other similar statements include trace, notice, warn, error and fatal. All of these statements accept arbitrary key-value pairs. As a short-cut, you are also allowed to specify only the name of a particular variable and Chronicles will create a key with the same name (i.e. passing a local variable named foo will be translated to the pair foo = foo).

A common practice enforced in other logging libraries is to associate the logging records with the name of the component that produced them or with a particular run-time property such as RequestID. Chronicles provides two general-purpose facilities for assigning such properties in an automated way:

logScope

logScope can be used to introduce additional properties that will be automatically attached to all logging statements in the current lexical scope:

logScope:
  # Lexical properties are typically assigned to a constant:
  topics = "rendering opengl"

  # But you can also assign an expression that will be
  # evaluated on every log statement:
  memoryUsage = currentMemUsage()

proc renderFrame(...) =
  inc frameCounter

  logScope:
    # You can add additional properties in any scope. Only logging
    # statements that are in the same lexical scope will be affected:
    frame = frameCounter

  var t = startTimer()
  debug "Frame started"

  ...

  glFinish()
  debug "Frame finished", totalPrimitives, frameTime = t.elapsed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23

A logScope is usually put near the top of a Nim module and used to specify statically assigned properties such as message origin, component name, etc. The special topics property demonstrated here is important for the log filtering mechanism, which will be explained in more details later. If present, this property will always appear first in the formatted log output.

publicLogScope

While a logScope affects only the current module, a publicLogScope allows you to specify a set of custom properties that may affect your entire program. For example, if you have an application running in a server cluster, you may want to assign a property such as serverId to every record. To achieve this, create a proxy logging module importing chronicles and setting up a publicLogScope:

# logging.nim

import chronicles

proc getServerId*()

publicLogScope:
  serverId = getServerId()

1
2
3
4
5
6
7
8
9

Every other module importing the proxy module will be able to use the entire Chronicles API and will be affected by the public scope. In fact, you should not import chronicles from such modules, because this will lead to ambiguous symbols such as activeChroniclesScope and activeChroniclesStream.

Using Nim's --import: option may be a good way to enforce the use of the proxy module in your entire program.

dynamicLogScope

A dynamicLogScope is a construct accepting a block of code that can be used to attach properties to all logging statements that will be executed anywhere within the tree of calls originating from the said block. The key difference with the lexically bound properties is that this includes logging statements from other modules, which are not within the lexical scope of the dynamicLogScope statement.

If you still find the distinction between lexical and dynamic scopes confusing, reading the following explanation may help you:

http://wiki.c2.com/?DynamicScoping

A dynamic scope is usually used to track the reason why a particular library function is being called (e.g. you are opening a file as a result of a particular network request):

proc onNewRequest(req: Request) =
  inc reqID
  info "request received", reqID, origin = req.remoteAddress
  dynamicLogScope(reqID):
    # All logging statements triggered before the current block returns
    # will feature the reqID property. This includes logging statements
    # from other modules.
    handleRequest(req)
1
2
3
4
5
6
7
8

Just like regular log statements, dynamicLogScope accepts a list of arbitrary key-value pairs. The use of reqID in the example above is a convenient short form for specifying the pair reqID = reqID.

While the properties associated with lexical scopes are lazily evaluated as previously demonstrated, all expressions at the beginning of a dynamic scope will be eagerly evaluated before the block is entered.

Compile-Time Configuration

Almost everything about Chronicles in configured at compile-time, through the mechanism of Nim's -d: flags. For example, you can completely remove all of the code related to logging by simply setting chronicles_enabled to off:

nim c -d:chronicles_enabled=off myprogram.nim
1

Chronicles comes with a very reasonable default configuration, but let's look at some of the other supported options:

chronicles_sinks

Chronicles supports producing log records in multiple formats and writing those to various destinations such as the std streams, the system's syslog daemon, or to one or more log files.

The combination of a log format and one or more associated log destinations is called a 'sink'. You can use the chronicles_sinks option to provide the list of sinks that will be used in your program.

The sinks are specified as a comma-separated list of valid Nim expressions that will be better illustrated by the following examples:

  • json

    Write JSON-records to stdout

  • json[file]

    Write JSON-records to a file in the current directory named after the application itself.

  • textblocks[stdout,file(/var/log/myapp.log)]

    Use the 'textblocks' format and send the output both to stdout and to a file with an absolute path /var/log/myapp.log

  • textlines[notimestamps,file(myapp.txt),syslog]

    Use the 'textlines' format, but don't include timestamps and write both to a file named 'myapp.txt' with a relative path to the current working directory and also to syslog.

  • textlines[nocolors],json[file(logs/myapp.json,truncate)]

    Send the output both in the 'textlines' format to stdout (but without using colors) and to a JSON file named myapp.json in the relative directory 'logs'. The myapp.json file will be truncated on each program execution.

The built-in formats include json, textlines and textblocks, which support options for specifying the use of colors and timestamps (for more info see chronicles_colors and chronicles_timestamps).

The possible log destinations are stdout, stderr, file and syslog.

Please note that Chronicles also allows you to implement custom logging formats through the use of the customLogStream facility.

chronicles_streams

While having multiple log sinks enables you to record the same stream of events in multiple formats and destinations, chronicles_streams allows you to define additional independent streams of events identified by their name. In the code, each logging statement is associated with exactly one log stream, which in turn has an associated list of sinks.

The syntax for defining streams closely resembles the syntax for defining sinks:

  • textlog[textlines],transactions[json[file(transactions.json)]]

This will create two streams, called textlog and transactions. The former will be considered the default stream associated with unqualified logging statements, but each of the streams will exist as a separate symbol in the code, supporting the full set of logging operations:

textlog.debug "about to create a transaction"
transactions.info "transaction created", buyer = alice, seller = bob
1
2

The streams created through chronicles_streams will be exported by the chronicles module itself, but you can also introduce additional streams in your own modules by using the helpers logStream and customLogStream.

chronicles_enabled_topics

All logging statements may be associated with a statically known list of topics. Usually, this is done by specifying the topics property in a particular logScope, but you can also specify it for individual log statements.

You can use the chronicles_enabled_topics option to specify the list of topics for which the logging statements should produce output. All other logging statements will be erased at compile-time from the final code. When the list includes multiple topics, any of them is considered a match.

In both contexts, the list of topics is written as a comma or space-separated string of case-sensitive topic names.

In the list of topics, you can also optionally provide a log level after the topic, separated with a colon from the topic. If a log level is provided it will overrule the chronicles_log_level setting. The log level can be defined as LogLevel values or directly as the corresponding integer values.

e.g. -d:chronicles_enabled_topics:MyTopic:DEBUG,AnotherTopic:5

chronicles_required_topics

Similar to chronicles_enabled_topics, but requires the logging statements to have all of the topics specified in this list.

You cannot specify chronicles_enabled_topics and chronicles_required_topics at the same time.

chronicles_disabled_topics

The dual of chronicles_enabled_topics. This option specifies a black-list of topics for which the associated logging statements should be erased from the program.

Topics in chronicles_disabled_topics have precedence over the ones in chronicles_enabled_topics or chronicles_required_topics.

chronicles_log_level

This option can be used to erase at compile-time all log statements, not matching the specified minimum log level.

Possible values are TRACE, DEBUG, INFO, NOTICE, WARN, ERROR, FATAL, and NONE. The default value is DEBUG in debug builds and INFO in release mode.

chronicles_runtime_filtering

This option enables the run-filtering capabilities of Chronicles. The run-time filtering is controlled through the procs setLogLevel and setTopicState:

type LogLevel = enum
  NONE, TRACE, DEBUG, INFO, NOTICE, WARN, ERROR, FATAL

proc setLogLevel*(level: LogLevel)

type TopicState = enum
  Normal, Enabled, Required, Disabled

proc setTopicState*(name: string,
                    newState: TopicState,
                    logLevel = LogLevel.NONE): bool
1
2
3
4
5
6
7
8
9
10
11

The log levels available at runtime - and therefor to setLogLevel() - are those greater than or equal to the one set at compile time by chronicles_log_level.

It is also possible for a specific topic to overrule the global LogLevel, set by setLogLevel, by setting the optional logLevel parameter in setTopicState to a valid LogLevel.

The option is disabled by default because we recommend filtering the log output in a tailing program. This allows you to still look at all logged events in case this becomes necessary. Set the option to on to enable it.

chronicles_timestamps

This option controls the use of timestamps in the log output. Possible values are:

  • RfcTime (used by default)

    Chronicles will use the human-readable format specified in RFC 3339: Date and Time on the Internet: Timestamps

    https://tools.ietf.org/html/rfc3339

  • UnixTime

    Chronicles will write a single float value for the number of seconds since the "Unix epoch"

    https://en.wikipedia.org/wiki/Unix_time

  • None or NoTimestamps

    Chronicles will not include timestamps in the log output.

Please note that the timestamp format can also be specified for individual sinks (see chronicles_sinks).

chronicles_line_numbers

This option, disabled by default, enables the display of filename and line number where each record was instantiated. It adds a property file to the output, for example:

file: example.nim:15
1

While chronicles_line_numbers sets the default option for all records, it is also possible to control the same property in a lexical scope or for a particular log statement with chroniclesLineNumbers, which can be either true or false.

chronicles_colors

This option controls the default color scheme used by Chronicles for its human-readable text formats when sent to the standard output streams.

Possible values are:

  • NativeColors (used by default)

    In this mode, Windows builds will produce output suitable for the console application in older versions of Windows. On Unix-like systems, ANSI codes are still used.

  • AnsiColors

    Output suitable for terminals supporting the standard ANSI escape codes: https://en.wikipedia.org/wiki/ANSI_escape_code

    This includes most terminal emulators on modern Unix-like systems, Windows console replacements such as ConEmu, and the native Console and PowerShell applications on Windows 10.

  • None or NoColors

    Chronicles will produce color-less output. Please note that this is the default mode for sinks logging only to files or for sinks using the json format.

Current known limitations:

  • Chronicles will not try to detect if the standard outputs of the program are being redirected to another program or a file. It's typical for the colored output to be disabled in such circumstances. (issue)

chronicles_indent

This option sets the desired number of spaces that Chronicles should use as indentation in the textblocks format.


All of the discussed options are case-insensitive and accept a number of truthy and falsy values such as on, off, true, false, 0, 1, yes, no or none.

Working with file outputs

When a stream has file outputs, you may choose to provide the log file location at run-time. Chronicles will create each log file lazily when the first log record is written. This gives you a chance to modify the default compile-time path associated with each file output by calling the open proc on an output symbol associated with the stream:

# my_program.nim

var config = loadConfiguration()
let success = defaultChroniclesStream.output.open(config.logFile, fmAppend)

info "APPLICATION STARTED"

1
2
3
4
5
6
7

Compiled with:

nim c -d:chronicles_sinks=textlines[file] my_program.nim
1

As you can see above, the default stream in Chronicles is called defaultChroniclesStream. If the stream had multiple file outputs, they would have been accessible separately as outputs[0], outputs[1] and so on. output is a simple short-cut referring to the first of them.

When the compile-time configuration doesn't specify a default file name for a particular file output, Chronicles will use the following rules for picking the default automatically:

  1. The log file is created in the current working directory and its name matches the name of the stream (plus a .log extension). The exception for this rule is the default stream, for which the log file will be assigned the name of the application binary.

  2. If more than one unnamed file outputs exist for a given stream, chronicles will add an index such as .2.log, .3.log .. .N.log to the final file name.

Teaching Chronicles about your types

Chronicles can output log records in any of the formats supported by the Nim serialization package. When you specify a named format such as json, Chronicles will expect that your project also depends on the respective serialization package (e.g. json_serialization).

In the text formats (textlines and textblocks), the Nim's standard $ operator will be used to convert the logged properties to strings.

formatIt

You can instruct Chronicles to alter this default behavior for a particular type by providing a chronicles.formatIt override:

type Dollar = distinct int
chronicles.formatIt(Dollar): "$" & $(it.int)
1
2

The formatIt block can evaluate to any expression that will be then subjected to the standard serialization logic described above.

expandIt

The expandIt override can be used to turn any logged property of a particular type into multiple properties:

chronicles.expandIt(EncryptedEnvelope):
  peer = it.fromAddress
  msg  = it.decryptMsg

...

var e = EncryptedEnvelope(...)

# The following two statements are equivalent:
info "Received message", e
info "Received message", peer = e.fromAddress, msg = e.decryptMsg
1
2
3
4
5
6
7
8
9
10
11

You can also derive the names of the expanded properties from the name of the original logged property. This is achieved by using the Nim's backticks syntax to construct the expanded property names:

chronicles.expandIt(User):
  # You can use both identifiers and string literals:
  `it Name` = it.name
  `it "LastSeen"` = it.lastSeen

...

var alice = User(name: "Alice", ...)

# The following two statements are equivalent:
info "Sending message", recipient = alice
info "Sending message", recipientName = alice.name, recipientLastSeen = alice.lastSeen
1
2
3
4
5
6
7
8
9
10
11
12

Custom Log Streams

logStream

As an alternative to specifying multiple output streams with the chronicles_streams option, you can also introduce additional streams within the code of your program. A typical way to do this would be to introduce a proxy module that imports and re-exports chronicles while adding additional streams with logStream:

import chronicles
export chronicles

logStream transactions[json[file(transactions.json)]]
1
2
3
4

The expression expected by logStream has exactly the same format as the compile-time option and produces the same effect. In this particular example, it will create a new stream called transactions that will be sent to a JSON file named transactions.json.

After importing the proxy module, you'll be able to create records with any of the logging statements in the usual way:

import transactions_log

...

transactions.error "payment gateway time-out", orderId,
                    networkStatus = obtainNetworkStatus()
1
2
3
4
5
6

customLogStream

customLogStream enables you to implement arbitrary log formats and destinations.

Each logging statement is translated to a set of calls operating over a structure called "Log Record" (with one instance created per logging statement). New log formats can be implemented by defining a suitable log record type. Let's demonstrate this by implementing a simple XML logger:

import xmltree, chronicles

type XmlRecord[Output] = object
  output: Output

template initLogRecord*(r: var XmlRecord, lvl: LogLevel,
                        topics: string, name: string) =
  r.output.append "<event name=\"", escape(name), "\" severity=\"", $lvl, "\">\n"

template setProperty*(r: var XmlRecord, key: string, val: auto) =
  r.output.append textBlockIndent, "<", key, ">", escape($val), "</", key, ">\n"

template setFirstProperty*(r: var XmlRecord, key: string, val: auto) =
  r.setProperty key, val

template flushRecord*(r: var XmlRecord) =
  r.output.append "</event>\n"
  r.output.flushOutput

customLogStream xmlout[XmlRecord[StdOutOutput]]

publicLogScope:
  stream = xmlout

info "New Video", franchise = "Tom & Jerry", episode = "Smarty Cat"
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

The produced output from the example will be:

<event name="New Video" severity="INFO">
  <tid>0</tid>
  <episode>Smarty Cat</episode>
  <franchise>Tom &amp; Jerry</franchise>
</event>
1
2
3
4
5

As you can see, customLogStream looks similar to a regular logStream, but it expects a log record type as its only argument.

The record type is implemented by providing suitable definitons for initLogRecord, setFirstProperty, setProperty and flushRecord. We recommend defining these operations as templates because this will facilitate the aggressive constant-folding employed by Chronicles (discussed in more details in the next section). We also recommend making your log record types parametric on an Output type, because this will allow the users of the code to specify any of the output types defined in Chronicles itself (see the module log_output for a list of those).

As demonstrated in the example above, you can set the stream property in a Chronicles lexical scope to redirect all unqualified log statements to a particular default stream.

Cost of Abstractions and Implementation Details

Chronicles makes use of advanced compile-time programming techniques to produce very efficient run-time code with minimal footprint.

The properties from lexical scopes are merged at compile-time with the log statement arguments and if any constant variables are about to be sent to the log output, they will be first concatenated by the compiler in order to issue the minimum number of write operations possible.

The dynamic scopes store their run-time bindings on the stack, in special frame structures forming a linked list. This list is traversed on each log statement and each active property leads to one dynamically dispatched call.

To support constant-time topic filtering and property overriding in dynamic scopes, Chronicles consumes a large amount of thread-local memory, roughly proportional to the number of unique topic names and property names used in the program.

Future Directions

At the moment, Chronicles intentionally omits certain features expected from a logging library such as log rotation and archival. We recommend following the guidelines set in the 12-factor app methodology and sending your log output to stdout. It should be the responsibility of the supervising daemon of the app to implement log rotation and archival.

We understand that certain users would want to take advantage of the file sinks provided by Chronicles and these users may benefit from the aforementioned features. If the Nim community provides a package for a low-level abstraction of an automatically rotated and archived log file, Chronicles will provide options for using it.