Case Study: Grammarly

“VictoriaMetrics Just Works - and Uses Fewer Hardware Resources Compared to Other Tools!”

Industry:
Internet
Location:
San Francisco, USA
Size:
500+  Employees

Grammarly’s digital writing assistant supports more than 30 million DAUs and 30,000 teams write more clearly and effectively every day. In building a product that scales across multiple platforms and devices, Grammarly works to empower users whenever and wherever they communicate.

Grammarly’s values-driven team is growing to support our expanding user base and to continue developing our writing assistant into a truly comprehensive communication partner. With a working model that balances remote work with in-person collaboration at Grammarly’s hubs in San Francisco, Kyiv, New York, and Vancouver, the Grammarly team strives to help people around the world connect and be understood.

Mission: Improve lives by improving communication.

Main Benefits of Using VictoriaMetrics

10x Cost Savings

Ingestion Types Flexibility

Performance

Easy to Get Started

Responsive VM Developers

Great Support & Docs

Challenge

The maintenance and scaling of our previous on-premises monitoring system was hard and required major engineering time and resources to maintain.

The stability of the previous solution was unreliable..

Our previous system struggled with storing frequently changing metrics (the moderate churn rate was a concern).

The overall costs of the previous solution were too high.

Solution

Ingestion type flexibility (support for Graphite, OpenMetrics, etc.) was definitely a winning feature and important benefit for us.

VictoriaMetrics comes with good documentation and is easy to bootstrap.

The high level of responsiveness of the VictoriaMetrics developers and support team during our research phase and production have made us extremely happy customers.

Delivered 10x cost savings versus our prior monitoring solution.

Why VictoriaMetrics Was Chosen Over Other Solutions

Great On-Premises Solution

Outperformed Competitive Solutions in Benchmarks

Strong POC Results

Direct Access & Great Support by VM Developers

After trying out SaaS solutions we decided to go with an in-house setup. Out of the various in-house tools we had short-listed, we decided to try VM first during a PoC taking into account publicly available benchmarking with competitive solutions. The PoC results and the VictoriaMetrics developers' help made it an easy decision to move forward with a VictoriaMetrics solution.

Technical Stats

Median memory usage during the last 24h

sum(avg_over_time(process_resident_memory_bytes[24h]))

618 GiB

The average number of cpu cores used during the last 24h

sum(rate(process_cpu_seconds_total[24h]))

~66 CPU cores

The maximum number of active time series during the last 24 hours

sum(max_over_time(vm_cache_entries{type="storage/hour_metric_ids"}[24h]))

~120 Mil

Daily time series churn rate

sum(increase(vm_new_timeseries_created_total[24h]))

~74 Mil

The average ingestion rate over the last 24h

sum(rate(vm_rows_inserted_total[24h]))

3.16Mil datapoints/sec

The total number of datapoints

sum(vm_rows{type=~"storage/.+"})

57.7 Tri

The total number of entries in inverted index

sum(vm_rows{type="indexdb"}))

295 Bil

Data size on disk

sum(vm_data_size_bytes{type=~"storage/.+"})

72.8 TiB

Index size on disk:

sum(vm_data_size_bytes{type="indexdb"})

7.6 TiB

The average datapoint size on disk

sum(vm_data_size_bytes) / sum(vm_rows)

~1.5 B

The average range query rate over the last 24h

sum(rate(vm_http_requests_total{path=~".*/api/v1/query_range"}[24h]))

~1.4 req/s

The average instant query rate over the last 24h

sum(rate(vm_http_requests_total{path=~".*/api/v1/query"}[24h]))

~47 req/s

Average range query duration quantiles over the last 24h

max(avg_over_time(vm_request_duration_seconds{path=~".*/api/v1/query_range"}[24h])) by (quantile)

1          2.29s
0.500  0.006s
0.900  0.3s
0.970  1.04s
0.990  2.21s

Average instant query duration quantiles over the last 24h

max(avg_over_time(vm_request_duration_seconds{path=~".*/api/v1/query"}[24h])) by (quantile)

1          19.8s
0.500  0.007s
0.900  0.2s
0.970  1s
0.990  2.6s