Troubleshooting
Fix Ignition Core Historian high CPU usage
Short answer
Gateway CPU above 80 percent that falls below 10 percent when only Core Historian is disabled is a historian storage fault, not a 30-tag acquisition load. Upgrade the 8.3.3 (b2026012009) gateway to 8.3.7 or newer, whose release notes fix Core Historian query performance degrading as data accumulated, and leave the partition interval and maintenance age as configured.
On this page
Yes, this is a Core Historian fault, and the fix is an upgrade, not a redesign of the tag setup. A gateway on Ignition 8.3.3 (b2026012009) whose CPU sits above 80 percent and drops below 10 percent the moment Core Historian is disabled, with projects, database connections and PLC connections still enabled, has already isolated the load. Thirty tags at one-second sampling are 30 evaluations per second, nowhere near enough to pin a gateway. The 8.3.7 release notes list a Core Historian fix for query performance that degraded as data accumulated, which matches a historian configured two to three months earlier that ran fine, took no new tags, and then slowed. Keep the partition interval at Hours and the maintenance age at 6 hours, upgrade, and verify against the same numbers.
Isolate the load with the historian-only toggle #
A gateway restart that changes nothing rules out a one-time event such as a startup burst or a garbage-collection cycle, and points at repeatable work. Disabling projects, database connections and PLC connections together can find a subsystem, but only a single-variable toggle proves it:
- Record the steady reading on the gateway performance dashboard with Core Historian enabled: above 80 percent.
- Disable Core Historian only, leaving every project, database connection and PLC connection enabled.
- Let the number settle and record it: below 10 percent.
- Re-enable the historian once. CPU climbing straight back completes the high-low-high proof.
- While the historian is disabled, confirm live tag values keep moving with good quality; that clears the controller, driver and acquisition paths.
The same controlled-disable method settles high CPU from any edge software, for example when troubleshooting high CPU usage in a Node-RED agent.
Worked example: 30 one-second tags against a 6-hour maintenance age #
Inputs, each from the installation in question:
- Historical tags: 30, from the tag history configuration.
- Sampling interval: 1 second, the configured evaluation frequency.
- Maintenance age: 6 hours, the Core Historian retention horizon.
- Partition interval: Hours, the Core Historian partitioning setting.
- Chart requirement: 30 minutes of trend.
- Observed gateway CPU with the historian enabled: above 80 percent; disabled: below 10 percent, both from the performance dashboard.
Rule in words: the nominal storage rate is the tag count times one sample per tag per second, and the store held at the retention horizon is that rate times the seconds in an hour times the hours, assuming every evaluation stores a value with no deadband suppression, which makes the total an upper bound.
Arithmetic:
- 30 tags x 1 sample/s = 30 samples/s
- 30 samples/s x 3,600 s/h = 108,000 samples/h
- 108,000 samples/h x 6 h = 648,000 samples on hand at the maintenance age
Result: 648,000 stored samples, with 30 evaluations per second arriving. CPU percentage does not convert into a sample capacity, so this sizes the store; it does not predict processor load.
Decision: 30 evaluations per second pinning a gateway, then releasing it the moment the historian is disabled, is abnormal at any core count worth checking. Carry the rule forward as a threshold: when the historian-only toggle moves CPU from above 80 percent to below 10 percent, treat the storage path as the fault and take the version branch. Retention work is finished.
Why the partition and retention changes did nothing #
The load returned at every partition interval and maintenance age tried, so partition sizing is not the branch. The settings also already do their job: a 6-hour maintenance age covers the 30-minute chart with 12x margin (6 h / 0.5 h = 12). Restoring the original Hours interval matters for a second reason: the 8.3.10 release notes fixed Core Historian partitions that were not properly archived after the partition scheme had been changed, so the interval changes made while troubleshooting are one more argument for landing on the current release rather than hunting for a better interval. If the store itself ever needs to shrink, deadband suppression on analog tags is the volume lever; the retention clock is not a CPU control.
Confirm the mechanism while CPU is high #
Diagnostics taken during the transition beat diagnostics taken minutes later. Mark the timestamp of the historian enable, the CPU rise and any clock-drift warning, then read the same window in the gateway logs: repeated errors or maintenance messages near the rise distinguish a failing loop from ordinary ingestion. Capture a thread dump while CPU is high. A historian thread repeatedly at the top supports the branch, heavy garbage collection points at allocation pressure, and a project script dominating the capture means coincidence, so review scripts that fire on history queries, chart refreshes or historian availability. Record the processor model, the host core count and the cores assigned to the gateway or its virtual machine, because one percentage can hide a single saturated execution path on a small allocation; compare per-core utilization where the host exposes it.
One design fact explains why this reads as gateway CPU: Core Historian runs inside the gateway process and always shares its processor, so there is nothing to relocate. Moving that load off the gateway means an external database historian, the trade-off weighed when evaluating Ignition historian with TimescaleDB for your plant.
The fix: upgrade from 8.3.3 to 8.3.7 or newer #
The affected build is 8.3.3 (b2026012009). The 8.3.7 release notes list, under Core Historian, a fix for an issue that caused Core Historian query performance to degrade as data accumulated. That is the documented defect closest to this signature, which is why the upgrade is the first corrective branch once the toggle test reproduces the fault. Go to 8.3.7 at minimum; the current 8.3.x release additionally carries the partition-archival fix described above. The verification below is what proves or displaces it.
Apply and verify #
- Export the gateway configuration per the site's recovery procedure, and run the change in a window where the parallel HMI remains the operators' authoritative view.
- Upgrade the gateway from 8.3.3 (b2026012009) to the current 8.3.x release, 8.3.7 at minimum.
- Restore the baseline exactly: 30 historical tags, 1-second sampling, partition interval Hours, maintenance age 6 hours. The version is the only variable that changes.
- Re-enable Core Historian and watch the dashboard through the first hours of accumulation, not just the first minutes.
- Pass: CPU holds well below the 80 percent line under the same load, and the 30-minute trend renders.
- Fail: the climb returns as data accumulates. Capture the running build, the logs around the rise including any clock-drift warnings, a thread dump taken while high, the core allocation and the exact historian configuration, and open one support case with the vendor.
Questions asked
Is 30 tags at one-second sampling too much for Core Historian?
No. That is 30 stored samples per second, 108,000 per hour and 648,000 across a 6-hour maintenance age, an upper bound assuming no deadband suppression. A gateway pinned above 80 percent on that rate has a storage-path fault, not a capacity problem; the historian-only disable test settles it.
Should I lower the maintenance age to reduce CPU?
No. Retention is not a CPU control here: the load returned at every partition interval and maintenance age tried, and 6 hours already covers the 30-minute chart with 12x margin. Restore the original settings and take the upgrade branch instead of trading away history.
Can I move Core Historian to its own server?
No. Core Historian runs inside the gateway process, so its work always lands on gateway CPU; there is no separate process to relocate. Moving to an external database historian is the architectural alternative when you want that load off the gateway.
Why did it run fine for months before the CPU climbed?
The 8.3.7 release notes describe Core Historian query performance degrading as data accumulated. The historian ran two to three months, took no new tags, then pinned the processor: accumulation, not configuration. That match is why the upgrade is the first branch and why the tag setup needs no redesign.
What do I capture if CPU stays high after upgrading?
The exact build now running, the gateway logs around the CPU rise including any clock-drift warnings, a thread dump taken while CPU is high, the host core count and the cores assigned to the gateway, and the exact historian settings. That package goes to the vendor support channel as one case.