Skip to main content

Scheduling Best Practices

Scheduling Best Practices

Configuration, monitoring, and optimization guide for scheduled jobs in Crestone.


Key Concepts

Crestone's scheduler allows you to define periodic executions of extractors, transformations, and data pipelines. Each job consumes server resources during execution; a poor configuration can lead to CPU contention, memory pressure, or database connection bottlenecks.

Scope of this guide

Applies to both Crestone deployment modes. Hardware metrics are primarily relevant for On-Premise environments. In SaaS, the infrastructure is managed by SEIDOR Analytics.


The tables below describe recommended thresholds by server load profile. CPU and RAM values are based on Crestone's minimum installation requirements. Aligning the scheduler's concurrency limits with these metrics prevents service degradation.

Server Profiles

ProfileCPU (cores)RAMDiskMax concurrent jobsTarget CPU usageTarget RAM usage
Minimum4 cores16 GB120 GB2–3≤ 60 %≤ 65 %
Recommended8 cores32 GB250 GB6–8≤ 70 %≤ 70 %
Production16 cores64 GB500 GB12–16≤ 75 %≤ 75 %
High availability32+ cores128+ GB1 TB+25+≤ 80 %≤ 80 %
Critical server threshold

If sustained CPU usage exceeds 85 % or RAM exceeds 90 %, Crestone may begin queuing jobs with delays or dropping them entirely. These thresholds must be monitored at the operating system level — not from within Crestone. Review your concurrency settings and scale the server if needed.

Quick Reference: Key Metrics

MetricValueDescription
CPU per job0.5–1 coreEstimated logical core per active job
RAM per job512 MBBaseline per extractor/transform
Free CPU headroom≥ 20 %Always reserve for peak spikes
Free RAM headroom≥ 25 %Prevent swap during load peaks

Job Time Distribution

  • ✅ Schedule high-cost jobs (large extractions, complex joins) during low-traffic windows: 02:00–06:00.
  • ✅ Avoid having more than 3 heavy jobs starting at the exact same minute.
  • ✅ Use staggered start times: if two jobs share the same frequency, offset their starts by at least 5 minutes.
  • ⚠️ Do not schedule production extractors in the same time slot as server-level database backups.
  • ⚠️ Avoid frequencies shorter than 5 minutes for jobs that run unindexed queries or full table scans.

Priorities and Queues

  • ✅ Assign high priority only to business-critical jobs (e.g. daily closes, executive dashboards).
  • ✅ Keep development and testing jobs separate from production to isolate failures and resource consumption.
  • ✅ In On-Premise environments with multiple sources, group jobs by origin or functional area to simplify operations and troubleshooting.

Common Mistakes to Avoid

Thundering herd

Occurs when many jobs with the same frequency (e.g. every hour on the hour) launch simultaneously. Staggering starts within 2–3 minute windows can significantly reduce CPU spikes on the server.

Implicit dependencies without order control

If job B consumes data produced by job A, make the dependency explicit in the scheduler. Running B before A produces inconsistent or empty data with no visible error in Crestone.

Unchecked execution history growth

Execution history grows indefinitely if not managed. We recommend keeping no more than 90 days of active history; archive or purge older records to avoid impacting the performance of Crestone's internal database.


Monitoring and Alerts

Crestone notifies the outcome of each job execution. For full environment observability, complement Crestone alerts with server-level monitoring tools.

Alerts available in Crestone

EventDescription
Job completed successfullySuccess notification sent when the execution finishes without errors.
Job completed with errorsFailure notification with details of the node or step that caused the error.

Metrics to monitor at the server level

The following metrics are not visible within Crestone and must be configured using OS or infrastructure tools (e.g. Prometheus, Grafana, Azure Monitor, CloudWatch):

MetricRecommended threshold
CPU usageAlert if sustained above 80 % for more than 15 minutes during active job execution.
Available RAMAlert if it drops below 2 GB at any point during active execution.
Disk spaceAlert if free space falls below 15 %, especially on the partition hosting Crestone's internal database.
Periodic review

Review your scheduling configuration monthly or whenever data volume changes significantly. The hardware thresholds above assume stable load; a 2× growth in data volume may require upgrading to the next server profile.