Specifying data types in value-pairs

By default, syslog-ng OSE handles every data as strings. However, certain destinations and data formats (for example, SQL, MongoDB, JSON , AMQP) support other types of data as well, for example, numbers or dates. The syslog-ng OSE application allows you to specify the data type in templates (this is also called type-hinting). If the destination driver supports data types, it converts the incoming data to the specified data type. For example, this allows you to store integer numbers as numbers in MongoDB, instead of strings.

From syslog-ng OSE version 4.0 onwards, name-value pairs are now triplets (name, type, value). Typing support is available for several other components, for example, json-parser() and the $(format-json) template function. For more information, see Components supported by data types.

CAUTION: Hazard of data loss!
If syslog-ng OSE cannot convert the data into the specified type, an error occurs, and syslog-ng OSE drops the message by default. To change how syslog-ng OSE handles data-conversion errors, see on-error().

To use type-hinting, enclose the macro or template containing the data with the type: <datatype>("<macro>"), for example: int("${PID}").

Currently the mongodb() destination and the format-json template function supports data types.

Example: Using type-hinting

The following example stores the MESSAGE, PID, DATE, and PROGRAM fields of a log message in a MongoDB database. The DATE and PID parts are stored as numbers instead of strings.

mongodb(
    value-pairs(
        pair("date", datetime("${UNIXTIME}"))
        pair("pid", int64("${PID}"))
        pair("program", "${PROGRAM}"))
        pair("message", "${MESSAGE}"))
    )
);

Use the following example to format the same fields into JSON.

$(format-json date=datetime("${UNIXTIME}") pid=int64("${PID}") program="${PROGRAM}" message="${MESSAGE}")

Use the following example to format the MESSAGE field as a JSON list.

$$(format-json message=list(${MESSAGE}))

The syslog-ng OSE application currently supports the following data-types.

• boolean: Converts the data to a boolean value. Anything that begins with a t or 1 is converted to true, anything that begins with an f or 0 is converted to false.

• datetime: Use it only with UNIX timestamps, anything else will likely result in an error. This means that currently you can use only the ${UNIXTIME} macro for this purpose.

• double: A floating-point number.

• literal: The data as a literal string, without adding any quotes or escape characters.

• int or int32: 32-bit integer.

• int64: 64-bit integer.

• string: The data is a string.

• list: The data is a list of strings.

• JSON: The data is a JSON snippet.

• null: The type of the data is undefined.

Components supported by data types

The following components support data types from syslog-ng OSE 4.0 and onwards:

NOTE: Component types not listed below process data as string.

• Numeric operators in filter expression comparisons are now type-aware. The exact comparison depends on the types associated with the values compared. For more information, see Comparing macro values in filters.

json-parser() and the format-json template function

For more information, see JSON parser and Template functions of syslog-ng OSE.

syslog-ng OSE converts all elements in a JSON object to name-value pairs, when using json-parser(). Any type related data present in the original JSON is retained. This data is propagated automatically to any other component that supports type, for example a destination.

Elements without type data are handled as strings.

JSON lists (arrays) are converted to syslog-ng OSE lists, and can be manipulated using the $(list-append) template functions.

set() and groupset() rewrite rules

The type of the field can be set. Type-casting can be executed using the set() and groupset() template functions, to properly promote the type information.

For more information, see Creating custom SDATA fields and Setting multiple message fields to specific values.

db-parser()

db-parser() rules can pair types with values using the type attribute.

Example: Using the type attribute

<value name="foobar" type="integer">${PID}</value>

The integer is a type-cast that couples $foobar with an integer type. The internal parsers of db-parser() (for example, @NUMBER@) automatically couple type information to the parsed name-value pair. For more information, see Using pattern databases.

add-contextual-data()

Name-value pairs that are populated using add-contextual-data() propagate type information, similarly to db-parser().

map-value-pairs()

map-value-pairs() propagates type information.

SQL type support

Columns with specific type information are stored with this information kept intact. For more information, see sql() destination options.

Template type support

Templates can be cast explicitly to a specific type. Templates also propagate type information from macros, template functions, and values in the template string.

python() typing

Python components (sources, destinations, parsers, and template functions) support all data types, except for json().

On-disk serialized formats (that is, disk buffer)

syslog-ng OSE Version 4.0 and newer versions are backwards compatible with messages serialized with earlier versions, and the format is compatible for downgrades. Therefore, even if a newer version of syslog-ng OSE serialized a message, older versions and associated tools are able to read it, however, in this case the type information is lost.

value-pairs()

Type:	parameter list of the value-pairs() option
Default:	empty string

Description: The value-pairs() option allows you to select specific information about a message easily using predefined macro groups. The selected information is represented as name-value pairs and can be used formatted to JSON format, or directly used in a mongodb() destination.

Example: Using the value-pairs() option

The following example selects every available information about the log message, except for the date-related macros (R_* and S_*), selects the ${.SDATA.meta.sequenceId} macro, and defines a new value-pair called MSGHDR that contains the program name and PID of the application that sent the log message.

value-pairs(
    scope(nv_pairs core syslog all_macros selected_macros everything)
    exclude("R_*")
    exclude("S_*")
    key(".SDATA.meta.sequenceId")
    pair("MSGHDR" "${PROGRAM}[${PID}]: ")
)

The following example selects the same information as the previous example, but converts it into JSON format.

$(format-json --scope nv_pairs,core,syslog,all_macros,selected_macros,everything \
    --exclude R_* --exclude S_* --key .SDATA.meta.sequenceId \
    --pair MSGHDR="${PROGRAM}[${PID}]: ")

NOTE: Every macro is included in the selection only once, but redundant information may appear if multiple macros include the same information (for example, including several date-related macros in the selection).

The value-pairs() option has the following parameters. The parameters are evaluated in the following order:

1. scope()
2. exclude()
3. key()
4. pair()

exclude()

Type:	Space-separated list of macros to remove from the selection created using the scope() option.
Default:	empty string

Description: This option removes the specified macros from the selection. Use it to remove unneeded macros selected using the scope() parameter.

For example, the following example removes the SDATA macros from the selection.

value-pairs(
    scope(rfc5424 selected_macros) 
    exclude(".SDATA*")       
)                            

The name of the macro to remove can include wildcards (*, ?). Regular expressions are not supported.

key()

Type:	Space-separated list of macros to be included in selection
Default:	empty string

Description: This option selects the specified macros. The selected macros will be included as MACRONAME=MACROVALUE, that is using key("HOST" will result in HOST = ${HOST}. You can use wildcards (*, ?) to select multiple macros.

For example:

value-pairs(                 
    scope(rfc3164)                                         
    key("HOST")                                            
)                                                          
                                                            
value-pairs(                                               
    scope(rfc3164)                                      
    key("HOST", "PROGRAM")                              
)                                                       

omit-empty-values()

Type:	flag
Default:	N/A

Description: If this option is specified, syslog-ng OSE does not include value-pairs with empty values in the output.

For example:

$(format-json --scope none --omit-empty-values)

or

value-pairs(                                           
    scope(rfc3164 selected-macros)                     
    omit-empty-values()                                
)                                                      

Available in syslog-ng OSE version 3.21 and later.

pair()

Type:	name value pairs in "<NAME>" "<VALUE>" format
Default:	empty string

Description: This option defines a new name-value pair to be included in the message. The value part can include macros, templates, and template functions as well.

For example:

value-pairs(                 
    scope(rfc3164)           
    pair("TIME" "${HOUR}:${MIN}") 
    pair("MSGHDR" "${PROGRAM}[${PID}]: ") 
)                            

rekey()

Type:	<pattern-to-select-names>, <list of transformations>
Default:	empty string

Description: This option allows you to manipulate and modify the name of the value-pairs. You can define transformations, which are are applied to the selected name-value pairs. The first parameter of the rekey() option is a glob pattern that selects the name-value pairs to modify. If you omit the pattern, the transformations are applied to every key of the scope. For details on globs, see Options of regular expressions. If you want to modify the names of several message fields, see also map-value-pairs: Rename value-pairs to normalize logs.

• If rekey() is used within a key() option, the name-value pairs specified in the glob of the key() option are transformed.
• If rekey() is used outside the key() option, every name-value pair of the scope() is transformed.

The following transformations are available:

• add-prefix("<my-prefix>")

  Adds the specified prefix to every name. For example, rekey(add-prefix("my-prefix."))

• replace-prefix("<prefix-to-replace>", "<new-prefix>")

  Replaces a substring at the beginning of the key with another string. Only prefixes can be replaced. For example, replace-prefix(".class",".patterndb") changes the beginning tag .class to .patterndb.
  This option was called replace() in syslog-ng OSE version 3.4.

• shift("<number>")

  Cuts the specified number of characters from the beginning of the name.

• lower

  Converts all keys to lowercase. Only characters included in the US ASCII are supported.

• shift-levels("<number>")

  Similar to --shift, but instead of cutting characters, it cuts dot-delimited "levels" in the name (including the initial dot). For example, --shift-levels 2 deletes the prefix up to the second dot in the name of the key: .iptables.SRC becomes SRC

• upper

  Converts all keys to uppercase. Only characters included in the US ASCII are supported.

Example: Using the rekey() option:

The following sample selects every value-pair that begins with .cee., deletes this prefix by cutting 4 characters from the names, and adds a new prefix(events.).

value-pairs(
    key(".cee.*"             
        rekey(               
            shift(4)         
                            
    add-prefix("events.") 
        )                    
    )                        
)                            

The rekey() option can be used with the format-json template-function as well, using the following syntax:

$(format-json --rekey .cee.* --add-prefix events.)

scope()

Type:	space-separated list of macro groups to include in selection
Default:	empty string

Description: This option selects predefined groups of macros. The following groups are available:

• nv-pairs: Every soft macro (name-value pair) associated with the message, except the ones that start with a dot (.) character. Macros starting with a dot character are generated within syslog-ng OSE and are not originally part of the message, therefore are not included in this group.

• dot-nv-pairs: Every soft macro (name-value pair) associated with the message which starts with a dot (.) character. For example, .classifier.rule_id and .sdata.*. Macros starting with a dot character are generated within syslog-ng OSE and are not originally part of the message.

• all-nv-pairs: Include every soft macro (name-value pair). Equivalent to using both nv-pairs and dot-nv-pairs.

• rfc3164: The macros that correspond to the RFC-3164 (legacy or BSD-syslog) message format: ${FACILITY}, ${PRIORITY}, ${HOST}, ${PROGRAM}, ${PID},${MESSAGE}, and ${DATE}.

• rfc5424: The macros that correspond to the RFC-5424 (IETF-syslog) message format: ${FACILITY}, ${PRIORITY}, ${HOST}, ${PROGRAM}, ${PID}, ${MESSAGE}, ${MSGID}, ${R_DATE}, and the metadata from the structured-data (SDATA) part of RFC-5424 formatted messages, that is, every macro that starts with .SDATA..

  The rfc5424 group also has the following alias: syslog-proto. Note that the value of ${R_DATE} will be listed under the DATE key. The rfc5424 group does not contain any metadata about the message, only information that was present in the original message. To include the most commonly used metadata (for example, the ${SOURCEIP} macro), use the selected-macros group instead.

• all-macros: Include every hard macro. This group is mainly useful for debugging, as it contains redundant information (for example, the date-related macros include the date-related information several times in various formats).

• selected-macros: Include the macros of the rfc3164 groups, and the most commonly used metadata about the log message: the ${TAGS}, ${SOURCEIP}, and ${SEQNUM} macros.

• sdata: The metadata from the structured-data (SDATA) part of RFC-5424 formatted messages, that is, every macro that starts with .SDATA.

• everything: Include every hard and soft macros. This group is mainly useful for debugging, as it contains redundant information (for example, the date-related macros include the date-related information several times in various formats).

• none: Reset previously added scopes, for example, to delete automatically-added name-value pairs. The following example deletes every value-pair from the scope, and adds only the ones starting with iptables: $(format-welf --scope none .iptables.*)

For example:

value-pairs(                 
    scope(rfc3164 selected-macros) 
)