Quick Start⌗

Victoria Metrics

  • An easy way to run VictoriaMetrics locally is to build from the source code as there’re lots of vm binaries like vmselect, vmstorage, vminsert, etc. to download, and all of them are maintained in the same repository, and Makefile is quite straightforward.
git clone git@github.com:VictoriaMetrics/VictoriaMetrics.git
make victoria-metrics
mv /bin/victoria-metrics /usr/bin
  • Run VictoriaMetrics With Systemd
[Unit]
Description="Victoria Metrics Single"
Documentation=https://https://docs.victoriametrics.com/
After=network.target

[Service]
Type=simple

ExecStart=/usr/bin/victoria-metrics-prod \
        -storageDataPath=/data/vm \
        -httpListenAddr=:8428 \
        -promscrape.config=/etc/prometheus/scrape.yml
ExecStop=/bin/kill -s SIGTERM $MAINPID

Restart=on-failure
SuccessExitStatus=0
LimitNOFILE=65536
StandardOutput=/var/log/vm/out.log
StandardError=/var/log/vm/err.log
SyslogIdentifier=prometheus

[Install]
WantedBy=multi-user.target

Storage⌗

On Disk Layout⌗

./data
├── big
│   ├── 2024_01
│   │   ├── 17A6101707C91B08
│   │   │   ├── index.bin
│   │   │   ├── metadata.json
│   │   │   ├── metaindex.bin
│   │   │   ├── timestamps.bin
│   │   │   └── values.bin
│   ├── 2024_02
│   ├── ...
│   └── snapshots
│       ├── 20240109092239-179E85E399D5EFDA
│       │   ├── 2023_12
│       │   └── 2024_01
│       └── 20240111104529-179E85E399D5EFDB
│           ├── 2023_12
│           └── 2024_01
├── flock.lock
└── small
    ├── 2024_01
    │   ├── 17A6101707CC58EF
    │   │   ├── index.bin
    │   │   ├── metadata.json
    │   │   ├── metaindex.bin
    │   │   ├── timestamps.bin
    │   │   └── values.bin
    └── snapshots
        ├── 20240109092239-179E85E399D5EFDA
        │   ├── 2023_12
        │   └── 2024_01
        └── 20240111104529-179E85E399D5EFDB
            ├── 2023_12
            └── 2024_01

For details, see doc

  • small:
  • big:
  • flock.lock:

for each part:

  • index.bin:
  • metadata.json:
  • metaindex.bin:
  • timestamps.bin:
  • values.bin:

snapshot:

Write Flow⌗

Ingested Data -> create reverted index(two type of index) -> Buffer -> In-memory Parts -> Disk (flush into disk periodically) -> Merge in the background -> Disk

When index created?

VM vs Prometheus (Disk) TBD

Read Flow⌗

Referrence⌗