Not seeing performance improvement with Victoria Metrics about victoriametrics

[Amper]

Hey @HarshitaJha.

In brief (but be sure to read the "Related links"):

• Shards are needed if you have too many scraping targets and have reached your vertical scaling limits. If you're migrating from a single Prometheus instance, you probably don't need sharding.
• If you really wish to achieve HA at the scraping level, then yes you need replication, but instead of cluster mode in vmagent you can just use HA pair - two equally configured vmagent replicas + deduplication at the database level. This is exactly how replicaCount works for vmagent in vm-operator. But that's up to you, the vmagent load doesn't last that long, so maybe losing a couple of scrape intervals on restart isn't a big deal for you.
• It all depends on what you want to achieve. Did you choose a cluster because you really need HA and unlimited scalability? In this case, it's not fair to compare performance (with the same amount of resources) because you've chose other benefits (HA, replication, per-component scalability) instead. If you don't need it, you may prefer Single-version, it will work faster with the same resources, see more details in this question.
  I would also like to point out that in its current form your cluster (replicationFactor=2 with 2 vmstorages) will not be able to support HA correctly If one storage node is unavailable, there is nowhere to reroute the second replica of the data (see Replication and data safety, Cluster availability and other related links).

For all other points I believe that you can find the answers in the attached "Related links".

Related links:

• vmagent:
  • Scraping big number of targets
  • vmagent High-Availability Examples
  • High availability (vmagent)
• vmcluster:
  • Which VictoriaMetrics type is recommended for use in production - single-node or cluster?
  • Cluster resizing and scalability
  • Capacity planning
  • Understand your setup size
  • Replication and data safety
  • Cluster availability
  • Slow data ingestion
  • Slow queries