prometheus-tsdb-dump reads a Prometheus TSDB block and writes metrics.
$ git clone https://github.com/ryotarai/prometheus-tsdb-dump.git
$ cd prometheus-tsdb-dump
$ make build
$ prometheus-tsdb-dump -block /path/to/prometheus-data/block-ulid -format victoriametrics
{"metric":{"__name__":"up","job":"node-exporter","instance":"a"},"values":[1,1,1],"timestamps":[1578636058619,1578636118619,1578636178619]}
{"metric":{"__name__":"up","job":"node-exporter","instance":"b"},"values":[1,1,1],"timestamps":[1578636058619,1578636118619,1578636178619]}
-format: Output format (default: victoriametrics)-min-timestamp: Minimum timestamp of exported samples (unix time in msec)-max-timestamp: Maximum timestamp of exported samples (unix time in msec)- The
-blockpath can point to a local directory or ans3://location. -dump-index: Dump block index information. The block path can point to a local directory or ans3://location.-aws-profile: AWS profile to use when accessing S3 for-dump-indexor when reading a block from S3 with-block-output: Write output to the given file instead of stdout-label-value: Comma-separated list of label values to filter by-label-key: Label name to apply with-label-value-metric-name: Dump only the series or index for the given metric name
S3 downloads will timeout after 5 minutes to avoid hanging operations. When reading blocks from S3 the index is streamed using ranged requests which reduces memory usage compared to downloading the entire file.
-metric-name can be used together with -label-key and -label-value to
filter by a specific metric and label value at the same time.
Output format can be configured via -format option.
victoriametrics format is a JSON streaming format which can be imported via VictoriaMetrics' /api/v1/import API. It looks like:
{"metric":{"__name__":"up","job":"node-exporter"},"values":[1,1,1],"timestamps":[1578636058619,1578636118619,1578636178619]}
First, take a snapshot of Prometheus TSDB:
$ curl -XPOST http://your-prometheus:9090/api/v1/admin/tsdb/snapshot
{"status":"success","data":{"name":"20200110T104512Z-xxxxxxxxxxxx"}}
Then, import data to VictoriaMetrics:
$ cd /path/to/prometheus/data/snapshots/20200110T104512Z-xxxxxxxxxxxx
$ parallelism="$(nproc)"
$ find . -mindepth 1 -maxdepth 1 -type d | xargs -n1 -P "$parallelism" sh -c 'echo $0; prometheus-tsdb-dump-linux -block "$0" -format victoriametrics | curl http://your-victoriametrics:8428/api/v1/import -T -'