Skip to content

Latest commit

 

History

History
282 lines (201 loc) · 7.51 KB

README.md

File metadata and controls

282 lines (201 loc) · 7.51 KB
Raw

Infix

Infix is an open source InfluxDB disk utility to manage and apply a set of rules to TSM and WAL files.

Infix works by scanning a storage directory for TSM (.tsm) and WAL (.wal) files and then applying a set of rules to data contained in those files.

Depending on rules being configured, Infix can rewrite TSM and WAL files to apply a set of transformations (rename a measurement, update a field's type, rename a field, ...)

Installation

To install Infix, run go get

go get -u github.com/Abc-Arbitrage/infix/command

Compile

You need golang >= 1.14

# Checkout Abc-Arbitrage/infix into your golang src directory:
cd golang/src/github.com/Abc-Arbitrage/infix/command
# Install package manager dep
dep ensure
go build -o infix

Usage

Usage: infix [options]

    -datadir
        Path to data storage (defaults to /var/lib/influxdb/data)
    -waldir
        Path to wal storage (defaults to /var/lib/influxdb/wal)
    -database
        The database to fix
    -retention
        The retention policy to fix
    -shard
        The id of the shard to fix
    -max-cache-size
        The maximum in-memory cache size in bytes (defaults to 1GB)
    -cache-snapshot-size
        The size in bytes after which the cache will be snapshotted to disk when re-writing TSM files (defaults to 25MB)
    -v
        Enable verbose logging
    -check
        Run in check mode (do not apply any change)
    -config
        The configuration file

Procedure

  • Stop InfluxDB

Before running infix, stop your influxd process

sudo systemctl stop influxdb
  • Run infix

Make sure to run infix with the appropriate user that owns your your TSM and WAL files.

sudo -u influxdb infix -datadir /var/lib/influxdb/data /var/lib/influxdb/wal -database telegraf -v -config rules.toml
  • Optional: rebuild the TSI index

If you configured infix to drop or rename measurements or series, make sure to rebuild your TSI index if you are using the tsi1 index type.

  • Restart InfluxDB

Restart InfluxDB by starting the influxd process

sudo systemctl start influxdb

Configuration

Rules and filters are configured in a TOML file.

This sections lists all the available rules as well as sample configuration

DropMeasurement Rule

This rule drops a given measurement

[[rules.drop-measurement]]
    [rules.drop-measurement.dropFilter.strings]
        hasprefix="linux."

will drop every measurement that starts with linux.

DropSerie Rule

This rules drops fields from a given serie

[[rules.drop-serie]]
    [rules.drop-serie.dropFilter.serie]
        [rules.drop-serie.dropFilter.serie.measurement.strings]
            equal="cpu"
        [rules.drop-serie.dropFilter.serie.tag.where]
            cpu="cpu0"
        [rule.drop-serie.dropFilter.serie.field.pattern]
            pattern="^(idle|usage_idle)$"

will drop values from fields idle and usage_idle from serie from cpu measurement with tag value cpu matching value cpu0 Note that field parameter can be omitted and all fields will be dropped.

OldSerie Rule

This rule identifies series with points older than a configured timestamp

[[rules.old-serie]]
    time="2020-01-01T00:08:00Z"
    out="stdout"
    #out="out_file.log"
    format="text"
    #format="json"
    #timestamp=true

will print series older than 2020-01-01 00:08:00 to stdout Output can be written to a file. Format can be either text or json. Setting timestamp to true will write the last timestamp to the output

RenameField Rule

This rules renames field from a given measurement

[[rules.rename-field]]
    to="agg_5m_${1}_${2}"
    [rules.rename-field.measurement.strings]
        hasprefix="linux."
    [rules.rename-field.field.pattern]
        pattern="(.+)_(avg|sum)"

will rename fields matching the pattern (.+)_(avg_sum) from measurements starting with linux. to agg_5m_${1}_${2}.

Note that if the field's filter is a pattern filter, the to can contains variables to replace matches of the regexp, expanded by golang's Regexp.ReplaceAll function

RenameTag Rule

This rules renames a tag from a given measurement

[[rules.rename-tag]]
    to="hostname"
    [measurement.strings]
        hasprefix="linux."
    [tag.strings]
        equal="host"

will rename tag host from measurements starting with the prefix linux. to hostname

RenameMeasurement Rule

This rules renames a measurement

[[rules.rename-measurement]]
    to="os"
    [rules.rename-measurement.from.pattern]
      pattern="^operating-system$"

will rename measurement operating-system to os.

UpdateFieldType Rule

This rule updates the type of a field from a given measurement

[[rules.update-field-type]]
    fromType="float"
    toType="integer"
    [rules.update-field.measurement.strings]
        equal="cpu"
    [rules.update-field.field.pattern]
        pattern="^(idle|active)"

will update the type of fields idle and active from the measurement cpu from float to integer

Filters

Most rules require filters. Filters are configured by dotted keys. Given a configuration parameter config-key and a filter filter-name, the corresponding key will be config-key.filter-name.

The fully qualified configuration key format for a filter applied to a given rule rule-name will be

[rules.rule-name.config-key.filter-name]
    # ... filter config

The section below lists all available filters and their configuration

PatternFilter

This filter filters keys based on pattern represented by golang Regexp class

    pattern="^(cpu|disk)$"

StringFilter

This filter filters keys based on golang strings package

    hasprefix="linux."
    hassuffix=".gauge"

will filter a key if it starts with linux. or ends with .gauge

Supported functions are

Note that functions are chained together with a or. The given configuration then translates to

strings.HasPrefix(key, "linux.") || strings.HasSuffix(key, ".gauge")

SerieFilter

This filter should be used with rules that act on series like DropSerie. This filter builds on 3 underlying filters:

  • Measurement to filter the measurement
  • Tag to filter tags
  • Field to filter fields

Measurement and Tag filters are mandatory. Field filter is optional. If none, all fields will pass the filter.

    [measurement.strings]
        equal="cpu"
    [tag.where]
        cpu="cpu0"
    [field.pattern]
        pattern="^(idle|usage_idle)$"

will filter series from measurement cpu (StringFilter) with tag cpu matching value cpu0 having fields idle or usage_idle.

WhereFilter

This filter is a special filter that can be used to filter tags and their corresponding values. It can be used in SerieFilter to filter tags.

    cpu="^(cpu0|cpu1)$"
    host="my-host"

will filter tag cpu matching pattern ^(cpu0|cpu1)$ and host matching my-host. Note that tag values can use patterns.