diff --git a/text/0097-log-data-model.md b/text/0097-log-data-model.md new file mode 100644 index 000000000..4bb155688 --- /dev/null +++ b/text/0097-log-data-model.md @@ -0,0 +1,952 @@ +# Log Data Model + +Introduce Data Model for Log Records as it is understood by OpenTelemetry. + +* [Motivation](#motivation) +* [Design Notes](#design-notes) + * [Requirements](#requirements) + * [Field Kinds](#field-kinds) +* [Logical Data Model vs Physical Format](#logical-data-model-vs-physical-format) +* [Log and Event Record Definition](#log-and-event-record-definition) + * [Field: `timestamp`](#field-timestamp) + * [Trace Context:](#trace-context) + * [Field: `trace_id`](#field-traceid) + * [Field: `span_id`](#field-spanid) + * [Field: `trace_flags`](#field-traceflags) + * [Severity](#severity) + * [Field: `severity_text`](#field-severitytext) + * [Field: `severity_number`](#field-severitynumber) + * [Displaying Severity](#displaying-severity) + * [Comparing Severity](#comparing-severity) + * [Field: `short_name`](#field-shortname) + * [Field: `body`](#field-body) + * [Field: `resource`](#field-resource) + * [Field: `attributes`](#field-attributes) +* [Representation](#representation) +* [Prior Art](#prior-art) + * [RFC5424 Syslog](#rfc5424-syslog) + * [Fluentd Forward Protocol Model](#fluentd-forward-protocol-model) +* [Appendix A. Example Mappings.](#appendix-a-example-mappings) + * [RFC5424 Syslog](#rfc5424-syslog-1) + * [Windows Event Log](#windows-event-log) + * [SignalFx Events](#signalfx-events) + * [Splunk HEC](#splunk-hec) + * [Log4j](#log4j) + * [Zap](#zap) + * [Apache](#apache) + * [CloudTrail Log Event](#cloudtrail-log-event) +* [Appendix B: `severity_number` example mappings.](#appendix-b-severitynumber-example-mappings) +* [Reference](#reference) + +## Motivation + +This is a proposal of a data model and semantic conventions that allow to +represent logs from various sources: application log files, machine generated +events, system logs, etc. Existing log formats can be unambiguously mapped to +this data model. Reverse mapping from this data model is also possible to the +extent that the target log format has equivalent capabilities. + +The purpose of the data model is to have a common understanding of what a log +record is, what data needs to be recorded, transferred, stored and interpreted +by a logging system. + +## Design Notes + +### Requirements + +This Data Model was designed to satisfy the following requirements: + +- It should be possible to unambiguously map existing log formats to this Data + Model. Translating log data from an arbitrary log format to this Data Model + and back should ideally result in identical data. + +- Mappings of other log formats to this Data Model should be semantically + meaningful. The Data Model must preserve the semantics of particular elements + of existing log formats. + +- Translating log data from an arbitrary log format A to this Data Model and + then translating from the Data Model to another log format B ideally must + result in a meaningful, lossless translation of log data that is no worse than + a reasonable direct translation from log format A to log format B. + +- It should be possible to efficiently represent the Data Model in concrete + implementations that require the data to be stored or transmitted. We + primarily care about 2 aspects of efficiency: CPU usage for + serialization/deserialization and space requirements in serialized form. This + is an indirect requirement that is affected by the specific representation of + the Data Model rather than the Data Model itself, but is still useful to keep + in mind. + +The data model aims to successfully represent 3 sorts of logs and events: + +- System Formats. These are logs and events generated by the operating system + and over which we have no control - we cannot change the format or affect what + information is included (unless the data is generated by an application which + we can modify). An example of system format is Syslog. + +- Third-party Applications. These are generated by third-party applications. We + may have certain control over what information is included, e.g. customize the + format. An example is Apache log file. + +- First-party Applications. These are applications that we develop and we have + some control over how the logs and events are generated and what information + we include in the logs. We can likely modify the source code of the + application if needed. + +### Field Kinds + +This data model defines a logical model for a log record (irrespective of the +physical format and encoding of the record). Each record contains 2 kinds of +fields: + +- Named top-level fields of specific type and meaning. + +- Fields stored in the key/value pair lists, which can contain arbitrary values + of different types. The keys and values for well-known fields follow semantic + conventions for key names and possible values that allow all parties that work + with the field to have the same interpretation of the data (see references to + semantic conventions for Resource and Attributes fields and examples in + Appendix A). + +The reasons for having these 2 kinds of fields is: + +- Ability to efficiently represent named top-level fields, which are almost + always present (e.g. when using encodings like Protocol Buffers where fields + are enumerated but not named on the wire), and to enforce their types and + domain of values, which is very useful for compiled languages with type + checks. + +- Flexibility to represent less frequent data via key/value pair lists. This + includes well-known data that has standardized semantics as well as arbitrary + custom data that the application may want to include in the logs. + +When designing this data model I followed the following reasoning to make a +decision about when to use use a top-level named field: + +- The field needs to be either mandatory for all records or be frequently + present in well-known log and event formats (such as timestamp) or is expected + to be often present in log records in upcoming logging systems (such as + trace_id). + +- The field’s semantics must be the same for all known log and event formats and + can be mapped directly and unambiguously to this data model. + +Both of the above conditions were required to give the field a place in the +top-level structure of the record. + +## Logical Data Model vs Physical Format + +The data model does not define the actual encoding and format of the log record +representation. Format definitions will be done in separate OTEPs (e.g the log +records may be represented as msgpack, JSON, Protocol Buffer messages, etc). + +## Log and Event Record Definition + +Note: below we use type any, which can be a scalar value (number, string or +boolean), or an array or map of values. Arbitrary deep nesting of values for +arrays and maps is allowed (essentially allow to represent an equivalent of a +JSON object). + +Appendix A contains many examples that show how existing log formats map to the +fields defined below. If there are questions about the meaning of the field +reviewing the examples may be helpful. + +Here is the list of fields in a log record: + +Field Name | +---------------| +timestamp | +trace_id | +span_id | +trace_flags | +severity_text | +severity_number| +short_name | +body | +resource | +attributes | + +Below is the detailed description of each field. + + +### Field: `timestamp` + +Type: Timestamp, uint64 nanosecods since Unix epoch + +Description: Time when the event occurred measured by the origin clock. This +field is optional, it may be missing the timestamp is unknown. + +### Trace Context: + +#### Field: `trace_id` + +Type: byte sequence + +Description: Optional request trace id. Can be set for logs that are part of +request processing and have an assigned trace id. + +#### Field: `span_id` + +Type: byte sequence + +Description: Optional span id. Can be set for logs that are part of a particular +processing span. If span_id is present trace_id SHOULD be also present. + +#### Field: `trace_flags` + +Type: byte + +Description: Optional trace flag as defined in W3C trace context specification. +At the time of writing the specification defines one flag - the SAMPLED flag. + +### Severity + +#### Field: `severity_text` + +Type: string + +Description: the severity text (also known as log level). This is an optional +field and is the original string representation as it is known at the source. If +this field is missing and `severity_number` is present then the short name that +corresponds to the `severity_number` can be used as a substitution. + +#### Field: `severity_number` + +Type: number + +Description: numerical value of the severity, normalized to values described in +this document. This is an optional field. If `severity_number` is missing and +severity_text is present then it may be assumed that `severity_number` is equal +to INFO (numeric 9) (see the meaning below). + +`severity_number` is an integer number. Smaller numerical values correspond to +less severe events (such as debug events), larger numerical values correspond to +more severe events (such as errors and critical events). The following table +defines the meaning of `severity_number` value: + +severity_number range|Range name|Meaning +---------------------|----------|------- +1-4 |TRACE |A fine-grained debugging event. Typically disabled in default configurations. +5-8 |DEBUG |A debugging event. Often is not emitted in default configurations. +9-12 |INFO |An informational event. Indicates that an event happened. +13-16 |WARN |A warning event. Not an error but is likely more important than an informational event. +17-20 |ERROR |An error event. Something went wrong. +21-24 |FATAL |A fatal error such as application or system crash. + +Smaller numerical values in each range represent less important (less severe) +events. Larger numerical values in each range represent more important (more +severe) events. For example `severity_number=17` describes an error that is less +critical than an error with `severity_number=20`. + +*Mapping of `severity_number`* + +Mappings from existing logging systems and formats (of source format for short) +must define how severity (or log level) of that particular format corresponds to +`severity_number` of this data model based on the meaning listed for each range in +the above table. + +If the source format has more than one severity that matches a single range in +this table then the severities of the source format must be assigned numerical +values from that range according to how severe (important) the source severity +is. + +For example if the source format defines "Error" and "Critical" as error events +and "Critical" is a more important and more severe situation then we can choose +the following `severity_number` values for the mapping: "Error"->17, +"Critical"->18. + +If the source format has only a single severity that matches the meaning of the +range then it is recommended to assign that severity the initial value of the +range. + +For example if the source format has an "Informational" log level and no other +log levels with similar meaning then it is recommended to use `severity_number=9` +for "Informational". + +Source formats that do not define a concept of severity or log level MAY omit +`severity_number` and severity_text fields. Backend and UIs may represent log +records with missing severity information distinctly or may interpret log +records with missing `severity_number` and `severity_text` fields as if the +`severity_number` was set equal to INFO (numeric value of 9). + +*Reverse Mapping* + +When performing a reverse mapping from `severity_number` to a specific format and +the `severity_number` has no corresponding mapping entry for that format then it +is recommended to choose the target severity that is in the same severity range +and is closest numerically. + +For example Zap has only one severity in the INFO range, called "Info". When +doing reverse mapping all values in INFO range (numeric 9-12) will be mapped to +Log4J’s "Info" level. + +*Error Semantics* + +If `severity_number` is present and has a value of ERROR (numeric 17) or higher +then it is an indication that the log record represents an erroneous situation. +It is up to the reader of this value to make a decision on how to use this fact +(e.g. UIs may display such errors in a different color or have a feature to find +all "errors"). + +If the log record represents an erroneous event and the source format does not +define a severity or log record field it is recommended to set severity_number +to ERROR (numeric 17) during the mapping process. If the log record represents a +non-erroneous event the `severity_number` field may be omitted or may be set to +any numeric value less than ERROR (numeric 17). The recommended value in this +case is INFO (numeric 9). See Appendix for more mapping examples. + +#### Displaying Severity + +The following table defines the recommended short name for each +`severity_number` value (this can be used for example for representing the +`severity_number` in the UI): + +severity_number|Short Name +---------------|---------- +1 |TRACE +2 |TRACE2 +3 |TRACE3 +4 |TRACE4 +5 |DEBUG +6 |DEBUG2 +7 |DEBUG3 +8 |DEBUG4 +9 |INFO +10 |INFO2 +11 |INFO3 +12 |INFO4 +13 |WARN +14 |WARN2 +15 |WARN3 +16 |WARN4 +17 |ERROR +18 |ERROR2 +19 |ERROR3 +20 |ERROR4 +21 |FATAL +22 |FATAL2 +23 |FATAL3 +24 |FATAL4 + +When an individual log record is displayed it is recommended to show both +severity_text and seveirty_number values. A recommended combined string in this +case begins with the short name followed by severity_text in parenthesis. + +For example "Informational" Syslog record will be displayed as INFO +(Informational). When for a particular log record the `severity_number` is +defined but the severity_text is missing it is recommended to only show the +short name, e.g. INFO. + +When drop down lists or other UI elements that are intended to represent the +possible set of values are used for representing the severity it is likely +preferable to use the short names. + +For example a dropdown list of severities that allows filtering log records by +severities is likely to be more usable if it contains the short names of +`severity_number` (and thus has a limited upper bound of elements) compared to a +dropdown list which lists all distinct severity_text values that are known to +the system (which can be a large number of elements, often differing only in +capitalization or abbreviated, e.g. "Info" vs "Information"). + +#### Comparing Severity + +In the contexts where severity participates less-than / greater-than comparisons +`severity_number` field should be used. `severity_number` can be compared to +another `severity_number` or to numbers in the 1..24 range (or to the +corresponding short names). + +When severity is used in equality or inequality comparisons (for example in +filters in the UIs) the recommendation is to attempt to use both severity_text +and short name of `severity_number` to perform matches. For example if we have a +record with severity_text field equal to "Informational" and `severity_number` +field equal to INFO then it may be preferable from the user experience +perspective to ensure that severity="Informational" and severity="INFO" +conditions both to are TRUE for that record. + +### Field: `short_name` + +Type: string + +Description: Short event identifier that does not contain varying parts. +`short_name` describes what happened (e.g. "ProcessStarted"). Recommended to be +no longer than 50 characters. Optional. Not guaranteed to be unique in any way. +Typically used for filtering and grouping purposes in backends. + +### Field: `body` + +Type: any + +Description: A value containing the body of the log record (see the description +of any type above). Can be for example a human-readable string message +(including multi-line) describing the event in a free form or it can be a +structured data composed of arrays and maps of other values. Can vary for each +occurrence of the event coming from the same source. + +### Field: `resource` + +Type: key/value pair list + +Description: Describes the source of the log, aka +[resource](https://github.com/open-telemetry/opentelemetry-specification/blob/master/specification/overview.md#resources). +"value" is of any type. Multiple occurrences of events coming from the same +event source can happen across time and they all have the same value of +resource. Can contain for example information about the application that emits +the record or about the infrastructure where the application runs. Data formats +that represent this data model may be designed in a manner that allows the +`resource` field to be recorded only once per batch of log records that come +from the same source. SHOULD follow OpenTelemetry +[semantic conventions](https://github.com/open-telemetry/opentelemetry-specification/tree/master/specification/resource/semantic_conventions) +for Resources. + +### Field: `attributes` + +Type: key/value pair list + +Description: Additional information about the specific event occurrence. "value" +is of any type. Unlike the resource field, which is fixed for a particular +source, attributes can vary for each occurrence of the event coming from the same +source. Can contain information about the request context (other than +TraceId/SpanId). SHOULD follow OpenTelemetry +[semantic conventions](https://github.com/open-telemetry/opentelemetry-specification/tree/master/specification/trace/semantic_conventions) +for Attributes. + +## Representation + +Representation of Log Data over the wire and the transmission protocol is to be +defined. + +Below are examples that show one possible representation in JSON. Note: this is +just an example to help understand the data model, don’t treat it as the way to +represent this data model in JSON, that is still to be defined. + +Example 1 + +```json +{ + "timestamp": 1586960586000, // JSON needs to make a decision about + // how to represent nanos. + "attributes": { + "http.status_code": 500, + "http.url": "http://example.com", + "my.custom.application.tag": "hello", + }, + "resource": { + "service.name": "donut_shop", + "service.version": "semver:2.0.0", + "k8s.pod.name": "1138528c-c36e-11e9-a1a7-42010a800198", + }, + "trace_id": "f4dbb3edd765f620", // this is a byte sequence + // (hex-encoded in JSON) + "span_id": "43222c2d51a7abe3", + "severity": "INFO", + "body": "20200415T072306-0700 INFO I like donuts" +} +``` + +Example 2 + +```json +{ + "timestamp": 1586960586000, + ... + "body": { + "i": "am", + "an": "event", + "of": { + "some": "complexity" + } + } +} +``` + +## Prior Art + +### RFC5424 Syslog + +RFC5424 defines structure log data format and protocol. The protocol is +ubiquitous (although unfortunately many implementations don’t follow structured +data recommendations). Here are some drawbacks that do not make Syslog a serious +contender for a data model: + +- While it allows structured attributes the body of the message can be only a + string. + +- Severity is hard-coded to 8 possible numeric values, and does not allow custom + severity labels. + +- Structured data does not allow arbitrary nesting and is 2-level only. + +- No clear separate place to specify data source (aka resource). There are a + couple hard-coded fields that serve this purpose in a limited way (HOSTNAME, + APP-NAME, FACILITY). + +### Fluentd Forward Protocol Model + +Forward protocol defines a log Entry concept as a timestamped record. The record +consists of 2 elements: a tag and a map of arbitrary key/value pairs. + +The model is universal enough to represent any log record. However, here are +some drawbacks: + +- All attributes of a record are represented via generic key/value pairs (except + tag and timestamp). This misses the optimization opportunities (see [Design + Notes](#design-notes)). + +- There is no clear separate place to specify data source (aka resource). + +- There is no mention of how exactly keys should be named and what are expected + values. This lack of any naming convention or standardization of key/value + pairs makes interoperability difficult. + + +## Appendix A. Example Mappings. + +This section contains examples of mapping of legacy events and logs formats to +the new universal data model. + +### RFC5424 Syslog + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
PropertyTypeDescriptionMaps to Unified Model Field
TIMESTAMPtimestampTime when an event occurred measured by the origin clock.timestamp
SEVERITYenumDefines the importance of the event. Example: `Debug`severity
FACILITYenumDescribes where the event originated. A predefined list of Unix processes. Part of event source identity. Example: `mail system`attributes["syslog.facility"]
VERSIONnumberMeta: protocol version, orthogonal to the event.attributes["syslog.version"]
HOSTNAMEstringDescribes the location where the event originated. Possible values are FQDN, IP address, etc.resource["host.hostname"]
APP-NAMEstringUser-defined app name. Part of event source identity.resource["service.name"]
PROCIDstringNot well defined. May be used as a meta field for protocol operation purposes or may be part of event source identity.attributes["syslog.procid"]
MSGIDstringDefines the type of the event. Part of event source identity. Example: "TCPIN"short_description
STRUCTURED-DATAarray of maps of string to stringA variety of use cases depending on the SDID: +Can describe event source identity +Can include data that describes particular occurence of the event. +Can be meta-information, e.g. quality of timestamp value.SDID origin.swVersion map to resource["service.version"] + +SDID origin.ip map to attribute[net.host.ip"] + +Rest of SDIDs -> attributes["syslog.*"]
MSGstringFree-form text message about the event. Typically human readable.body
+ + +### Windows Event Log + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
PropertyTypeDescriptionMaps to Unified Model Field
TimeCreatedtimestampThe time stamp that identifies when the event was logged.timestamp
LevelenumContains the severity level of the event.severity
ComputerstringThe name of the computer on which the event occurred.resource["host.hostname"]
EventIDuintThe identifier that the provider used to identify the event.short_description
MessagestringThe message string.body
Rest of the fields.anyAll other fields in the event.attributes["winlog.*"]
+ + +### SignalFx Events + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
FieldTypeDescriptionMaps to Unified Model Field
TimestamptimestampTime when the event occurred measured by the origin clock.timestamp
EventTypestringShort machine understandable string describing the event type. SignalFx specific concept. Non-namespaced. Example: k8s Event Reason field.short_description
CategoryenumDescribes where the event originated and why. SignalFx specific concept. Example: AGENT. attributes["com.splunk.signalfx.event_category"]
Dimensionsmap of string to stringHelps to define the identity of the event source together with EventType and Category. Multiple occurrences of events coming from the same event source can happen across time and they all have the value of Dimensions. resource
Propertiesmap of string to anyAdditional information about the specific event occurrence. Unlike Dimensions which are fixed for a particular event source, Properties can have different values for each occurence of the event coming from the same event source.attributes
+ + +### Splunk HEC + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
FieldTypeDescriptionMaps to Unified Model Field
timenumeric, stringThe event time in epoch time format, in seconds.timestamp
hoststringThe host value to assign to the event data. This is typically the host name of the client that you are sending data from.resource["host.hostname"]
sourcestringThe source value to assign to the event data. For example, if you are sending data from an app you are developing, you could set this key to the name of the app.resource["service.name"]
sourcetypestringThe sourcetype value to assign to the event data.attributes["source.type"]
eventanyThe JSON representation of the raw body of the event. It can be a string, number, string array, number array, JSON object, or a JSON array.body
fieldsMap of anySpecifies a JSON object that contains explicit custom fields.attributes
indexstringThe name of the index by which the event data is to be indexed. The index you specify here must be within the list of allowed indexes if the token has the indexes parameter set.TBD, most like will go to attributes
+ + +### Log4j + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
FieldTypeDescriptionMaps to Unified Model Field
InstanttimestampTime when an event occurred measured by the origin clock.timestamp
LevelenumLog level.severity
MessagestringHuman readable message.body
All other fieldsanyStructured data.attributes
+ + +### Zap + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
FieldTypeDescriptionMaps to Unified Model Field
tstimestampTime when an event occurred measured by the origin clock.timestamp
levelenumLogging level.severity
callerstringCalling function's filename and line number. +attributes, key=TBD
msgstringHuman readable message.body
All other fieldsanyStructured data.attributes
+ + +### Apache + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
FieldTypeDescriptionMaps to Unified Model Field
%ttimestampTime when an event occurred measured by the origin clock.timestamp
%astringClient IPattributes["net.peer.ip"]
%AstringServer IPattributes["net.host.ip"]
%hstringRemote hostname. attributes["net.peer.name"]
%mstringThe request method.attributes["http.method"]
%v,%p,%U,%qstringMultiple fields that can be composed into URL.attributes["http.url"]
%>sstringResponse status.attributes["http.status_code"]
All other fieldsanyStructured data.attributes, key=TBD
+ + +### CloudTrail Log Event + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
FieldTypeDescriptionMaps to Unified Model Field
eventTimestringThe date and time the request was made, in coordinated universal time (UTC).timestamp
eventSourcestringThe service that the request was made to. This name is typically a short form of the service name without spaces plus .amazonaws.com.resource["service.name"]?
awsRegionstringThe AWS region that the request was made to, such as us-east-2.resource["cloud.region"]
sourceIPAddressstringThe IP address that the request was made from.resource["net.peer.ip"] or resource["net.host.ip"]? TBD
errorCodestringThe AWS service error if the request returns an error.short_description
errorMessagestringIf the request returns an error, the description of the error.body
All other fields*attributes["cloudtrail.*"]
+ + +## Appendix B: `severity_number` example mappings. + +Syslog |WinEvtLog |Log4j |Zap |severity_number +-------------|-----------|------|------|--------------- +- |- |TRACE |- |TRACE +Debug |Verbose |DEBUG |Debug |DEBUG +Informational|Information|INFO |Info |INFO +Notice | | | |INFO2 +Warning |Warning |WARN |Warn |WARN +Error |Error |ERROR |Error |ERROR +Critical |Critical |- |Dpanic|ERROR2 +- |- |- |Panic |ERROR3 +Alert |- |FATAL |Fatal |FATAL + +## Reference + +- Draft discussion of Data Model: + https://docs.google.com/document/d/1ix9_4TQO3o-qyeyNhcOmqAc1MTyr-wnXxxsdWgCMn9c/edit# + +- Example mappings of existing log formats to this Data Model: + https://docs.google.com/document/d/1ix9_4TQO3o-qyeyNhcOmqAc1MTyr-wnXxxsdWgCMn9c/edit?ts=5e990fe2#heading=h.ud60xroz7j2n + +- Discussion of Severity field: + https://docs.google.com/document/d/1WQDz1jF0yKBXe3OibXWfy3g6lor9SvjZ4xT-8uuDCiA/edit# \ No newline at end of file