Scaling and Sizing Guide

This guide provides sizing recommendations for RTFS deployments based on the number of Federated repositories, event volume, and member count.

Deployment Tiers

Suitable for teams getting started with Federation or running a limited number of synchronized repositories.

Suitable for organizations with multiple teams and moderate artifact volume.

Resource	Recommended Value
RTFS replicas	1 - 2
CPU requests and limits	4 and 6
Memory requests and limits	8Gi and 12Gi
Database	4 vCPU, 8 GB RAM
Thread pool overrides	Consider increasing `rtfs.jobs.thread.pool.size` and `rtfs.tenant.thread.pool.size`

Suitable for large enterprises with high artifact volume and many Federation members.

Resource	Recommended Value
RTFS replicas	2+ (with autoscaling)
CPU requests and limits	4 and 6 (per replica)
Memory requests and limits	8Gi and 12Gi (per replica)
Database	8+ vCPU, 16+ GB RAM
Thread pool overrides	Increase thread pools proportionally (see the following section)

Scale thread pools based on the number of Federated repositories and members.

Property	Default	When to Increase	Guideline
`rtfs.jobs.thread.pool.size`	100	> 100 repos or high Full Sync frequency	Set to 1.5x the number of federated repos
`rtfs.tenant.thread.pool.size`	100	Multi-tenant SaaS deployments	Scale with tenant count
`rtfs.heartbeat.thread.pool.size`	100	> 100 repos with many members each	Scale with total member count across all repos
`rtfs.tenant.heartbeat.thread.pool.size`	200	Multi-tenant deployments	Scale with tenant count
`rtfs.full.sync.thread.pool.size`	30	Frequent Full Syncs or large repos	Increase to handle concurrent Full Syncs
`rtfs.request.tunneling.thread.pool.size`	30	Heavy use of cross-member operations	Scale with tunneling request volume

The following properties control worker and batch sizing.

Property	Default	Impact of Increasing	Impact of Decreasing
`rtfs.worker.batch.size`	100	Higher throughput per worker cycle, more memory usage	Lower throughput, faster per-event response
`rtfs.worker.max.events.processed.in.batch`	200	More events per cycle for STANDARD priority repos	Limits maximum batch size
`rtfs.worker.max.events.processed.in.batch.for.high.priority`	400	More events per cycle for HIGH or URGENT repos	Limits maximum batch size for prioritized repos
`rtfs.stream.allocation.size`	50	More events loaded per stream read	More memory per stream

RTFS enforces limits on the number of repositories that can be set to HIGH or URGENT priority to prevent priority inflation.

Property	Default	Description
`rtfs.max.high.priority.repos.per.member`	10	Maximum number of HIGH priority repos per member.
`rtfs.max.high.priority.repos.percentage`	0.2 (20%)	Maximum percentage of total repos that can be HIGH priority.

The effective limit is the lower of the two values.

The following properties control connection pool sizing.

Property	Default	When to Increase
`rtfs.reactive.http.client.max.connections`	50	High concurrency across many members
`federated.repo.max.total.http.connections`	50	Many concurrent federation operations on Artifactory side

The following properties control binary task handler sizing.

Property	Default	When to Increase
`rtfs.binary.tasks.handler.tenants.thread.pool.size`	100	High volume of binary transfers
`rtfs.binary.tasks.handler.fetch.limit.per.tenant`	500	Large backlog of pending binary tasks
`rtfs.binary.tasks.inbound.insert.batch.size`	500	Burst of new binary tasks during Full Sync

RTFS supports Horizontal Pod Autoscaling when deployed via Helm:

autoscaling:
  enabled: true
  minReplicas: 1
  maxReplicas: 5
  targetCPU: "70"
  targetMemory: "80"

Use these metrics to determine if scaling is needed.

Metric	What It Tells You	Action If High
`jfrtfs_worker_lag`	Event processing is falling behind	Increase worker batch size or add replicas
`jfrtfs_binary_tasks_counter` (status=AVAILABLE)	Binary tasks are queuing up	Increase binary task handler thread pool
`jfrtfs_http_client_connections_pending`	HTTP connections are exhausted	Increase max connections
`jfrtfs_http_client_connections_available`	Low availability indicates saturation	Increase max connections or add replicas