FileWriter
Read the API documentation »
Output FileWriter
Overview
This is a raw output, for each record it receives a buffer containing the serialized record, plus a list holding a set of fields (output.fields
in TOML).
This output writes serialized records into compressed files, gzip (.gz) or zstd (.zst) depending on the file extension in PathString.
Generated files may be rotated if RotateInterval is set. PathString is used to control the name of the generated files, it may contain placeholders. These placeholders are evaluated each time a file is created, that is upon creation of the output or everytime a rotation takes place.
Supported placeholders:
- {{.Year}} year at file creation, 4 digits (YYYY)
- {{.Month}} month number at file creation, 2 digits (MM)
- {{.Day}} day of the month at file creation, 2 digits (DD)
- {{.Hour}} hour at file creation in 24h format, 2 digits (HH)
- {{.Minute}} minute at file creation, 2 digits (MM)
- {{.Second}} second at file creation, 2 digits (SS)
- {{.Index}} index of the current output process (see [output.procs]), 4 digits long
- {{.UUID}} per-worker random UUID (v4 UUID), 36 chars long
- {{.Rotation}} rotation count, 6 digits long
- {{.Field0}} value of the first field provided in [output.fields] (only if present).
When choosing configuration values for your FileWriter, it’s important to keep in mind the following rules:
- A file should only ever be accessed by a single worker at a time.
If you use multiple output processes, you should use {{.Index}} or {{.UUID}} so that generated filenames are guaranteed to be different for each workers.
- Rotation should never generate the same path twice.
To avoid a file to be overwritten by its successor in the rotation, you should ensure that 2 files generated at a distance of RotateInterval will have different filenames. To ensure filenames are different, you should set RotateInterval to a duration that exceeds that of the time-based placeholder with the shortest span.
For example, the following is correct since it’s the generated path is guaranteed to be unique at each rotation:
PathString = "/path/to/file-{{.Hour}}-{{.Minute}}.log.gz"
RotateInterval = 5m
However, this is not correct, since successive generations may generate the exact same path:
PathString = "/path/to/file-{{.Hour}}-{{.Minute}}.log.gz"
RotateInterval = 1s
If you use RotateSize you should also probably include {{.Rotation}}, to generate different different paths:
PathString = "/path/to/file-{{.Rotation}}.log.gz"
RotateSize = "128MB"
- Only use {{.Field0}} if you trust the records you consume.
By using {{.Field0}} the files produces will have a path containing whatever value is found. It could contain characters that are not valid to appear in a path. That also means that the number of files (and workers) depend on the cardinality of that field.
Configuration
Keys available in the [output.config]
section:
Name | Type | Default | Required | Description |
---|---|---|---|---|
PathString | string | "" | false | Template describing names of the generated files. See top-level documentation for supported placeholders.. |
RotateInterval | duration | 60s | false | Time interval between 2 successive file rotations. -1 disables interval-based rotation. |
RotateSize | bytes as int or string with SI or IEC unit | 0 | false | File size which when reached triggers a file rotation. Can be cumulated with RotateInterval. 0 to disable. Examples: 12000, 12KB, 1MB, 1MiB, etc. |
DiscardEmptyFiles | bool | false | false | By default, rotation may create empty files if no records were received. If true, then rotation is skipped if the file would be empty. |
ZstdCompressionLevel | int | 3 | false | Zstd compression level, ranging from 1 (best speed) to 19 (best compression). |
ZstdWindowLog | int | 0 | false | Enable zstd long distance matching. Increase memory usage for both compressor/decompressor. If more than 27 the decompressor requires special treatment. 0:disabled. |