Using Vector to ingest logs into Uptrace

Vector collects, transforms, and sends your logs to multiple destinations including Uptrace. It is blazingly fast and memory efficient.

Configuration

To configure Vector to send logs to Uptrace, use the HTTP sink and pass your project DSN via HTTP headers.

For example, to collect syslog messages, you can create the following Vector config:

toml

[sources.syslog]
type = "file"
include = ["/var/log/syslog"]

[transforms.parse_syslog]
type = "remap"
inputs = ["syslog"]
source = '''
. = parse_syslog!(string!(.message))
'''

[sinks.uptrace]
type = "http"
method = "post"
inputs = ["parse_syslog"]
encoding.codec = "json"
framing.method = "newline_delimited"
compression = "gzip"
request.headers.uptrace-dsn = "<FIXME>"
uri = "https://api.uptrace.dev/api/v1/vector/logs"

Copy the config above to vector.toml and then start Vector:

shell

vector --config=vector.toml

To see the data Vector sends to Uptrace, use the console sink:

toml

[sinks.my_sink_id]
type = "console"
inputs = [ "my-source-or-transform-id" ]

See vector-logs example for details.

VRL and attributes

Vector remap language (VRL) allows you to parse logs and set key-value pairs (attributes), for example:

toml

[transforms.parse_apache]
type = "remap"
inputs = ["apache_common_logs"]
source = '''
. = parse_apache_log!(string!(.message), "common")
.log_source = "apache"
'''

You should be careful with attribute names that contain a dot, because Vector uses the dot to create nested structures. For example, this Vector statement:

toml

.service.name = "my_service"

Produces the following nested JSON, which Uptrace no longer recognizes as a service.name attribute and leaves as is:

json

{
  "service": {
    "name": "my_service"
  }
}

Instead, you should replace dots with underscores or quote the attribute name:

toml

.service_name = "my_service"
# or
."service.name" = "my_service"

Nested JSON

If you have nested JSON values, you might want to use flatten function to transform the value into a single-level representation, for example:

toml

. = flatten({
    "parent1": {
        "child1": 1,
        "child2": 2
    },
    "parent2": {
        "child3": 3
    }
})

Produces:

json

{
  "parent1.child1": 1,
  "parent1.child2": 2,
  "parent2.child3": 3
}

Display name and grouping

Typically Uptrace is able to automatically generate a short summary for logs using the log.message attribute, but occasionally you may want to provide a custom summary.

This is where the display_name attribute comes in handy. The display_name attribute is a human-readable string that provides a short summary of the log event.

toml

[transforms.parse_apache]
type = "remap"
inputs = ["apache_common_logs"]
source = '''
. = parse_apache_log!(string!(.message), "common")
.display_name = join([.protocol, .method], " ") ?? ""
'''

Uptrace does not use display names for grouping, so you're free to put whatever you want there.

To control how Uptrace groups logs together, you can specify the grouping_fingerprint attribute which can be a string or a number (hash). Uptrace will group logs with the same fingerprint together.

toml

[transforms.parse_apache]
type = "remap"
inputs = ["apache_common_logs"]
source = '''
. = parse_apache_log!(string!(.message), "common")
.grouping_fingerprint = join([.protocol, .method], " ") ?? ""
'''

Log time

Uptraces uses the following fields in order to determine the log time:

timestamp
datetime
time

If such a field is found and it contains a valid time, Uptrace uses the time as the log time and drops the field.

For example:

text

timestamp=2006-01-02T15:04:05Z07:00 rest of the message
time="2006-01-02 15:04:05" rest of the message
{"datetime": "2006-01-02T15:04:05Z07:00", "message": "rest of the message"}

Uptrace Enterprise (including Uptrace Cloud) drops logs that are older than 24 hours. The same limitation applies to spans and metrics.

Converting logs to spans

When parsing HTTP or SQL logs, it might be useful to convert logs into spans by providing span name, duration, and some other fields.

For example, you can set attributes using remap transformation:

toml

[transforms.span_attrs]
type = "remap"
inputs = ["in"]
source = '''
.span_name = "<span name>"
.span_kind = "server"
.span_duration = 12345 # nanoseconds
'''

Uptrace recognizes the following span-related attributes:

trace_id in hex-encoded format, for example, 958180131ddde684c1dbda1aeacf51d3.
span_id is the id of the parent span, for example, 0cf859e4f7510204.
span_name is the span name. Required.
span_kind is the span kind.
span_duration is the span duration in nanoseconds.

You can also use all available semantic attributes.

Heroku logs

You can collect logs from Heroku’s Logplex and push them to Uptrace.

Fly logs

You can ship logs from Fly.io apps to Uptrace using NATS and Vector.