apache/skywalking · error · IllegalArgumentException

Incompatible ranks size = [{value.getRanks().length}] for cu

Error message

Incompatible ranks size = [{value.getRanks().length}] for current PercentileFunction[{ranks.size()}]

What it means

Once an AvgHistogramPercentileFunction instance has accepted a rank set (ranks is non-empty), every later accept() must supply the exact same rank count — the persisted 'ranks' column and percentile-reading logic are fixed to that set. A differing length throws 'Incompatible ranks size' (and a differing set of equal size throws the adjacent 'Rank ... doesn't exist' error).

Source

Thrown at oap-server/server-core/src/main/java/org/apache/skywalking/oap/server/core/analysis/meter/function/avg/AvgHistogramPercentileFunction.java:134

    @Override
    public void accept(final MeterEntity entity, final PercentileArgument value) {
        if (dataset.size() > 0) {
            if (!value.getBucketedValues().isCompatible(dataset)) {
                throw new IllegalArgumentException(
                    "Incompatible BucketedValues [" + value + "] for current PercentileFunction[" + dataset + "]");
            }
        }

        for (final int rank : value.getRanks()) {
            if (rank <= 0) {
                throw new IllegalArgumentException("Illegal rank value " + rank + ", must be positive");
            }
        }

        if (ranks.size() > 0) {
            if (ranks.size() != value.getRanks().length) {
                throw new IllegalArgumentException(
                    "Incompatible ranks size = [" + value.getRanks().length + "] for current PercentileFunction[" + ranks
                        .size() + "]");
            } else {
                for (final int rank : value.getRanks()) {
                    if (!ranks.include(rank)) {
                        throw new IllegalArgumentException(
                            "Rank " + rank + " doesn't exist in the previous ranks " + ranks);
                    }
                }
            }
        } else {
            for (final int rank : value.getRanks()) {
                ranks.add(rank);
            }
        }

        this.entityId = entity.id();

View on GitHub (pinned to 102af09b4a)

Solutions

Keep the rank list fixed for the lifetime of the percentile metric name — change ranks only together with a new metric name
If ranks must change, hot-remove the old metric / restart OAP so the function restarts with an empty ranks state
Search all rules for the metric name and make the rank lists identical

Example fix

# before — hot-edit adds a rank to an existing metric
  percentiles: 50,99      # original
  percentiles: 50,90,99   # edited -> Incompatible ranks size = [3] ... [2]

# after — new metric name for the new rank set
  metric: service_latency_percentile_v2
  percentiles: 50,90,99

Defensive patterns

Strategy: validation

Validate before calling

// keep the rank set constant per metric name; validate against the contract before accept
if (!Arrays.equals(ranks, rankContractFor(metricName))) {
    log.warn("Rank set changed for {} — use a new metric name or reset state", metricName);
    return;
}
percentileMetric.accept(entity, new PercentileArgument(bucketedValues, ranks));

Try / catch

try {
    percentileMetric.accept(entity, percentileArgument);
} catch (IllegalArgumentException e) {
    if (e.getMessage().contains("Incompatible ranks size")
            || e.getMessage().contains("doesn't exist in the previous ranks")) {
        log.error("Rank drift on {} — ranks are immutable once accepted", metricName, e);
        return;
    }
    throw e;
}

Prevention

Consider ranks part of the metric identity: to change them, ship a new metric name
If ranks must change on an existing name, hot-remove the metric or restart OAP so ranks state resets
Define the percentile list once in shared config and reuse it in every rule referencing the metric

When it happens

Trigger: A MAL percentile rule is edited to add or remove a rank (e.g. 50,99 -> 50,90,99) and re-applied while the metric's in-memory/persisted aggregation still carries the old rank set; two rules emit the same percentile metric with different rank lists.

Common situations: Hot-reloading percentile rules after editing the percentiles list; copy-pasted metric names across rules with diverging rank configs; upgrading a config that added a default rank.

Related errors

AI-assisted analysis of apache/skywalking@102af09b4a (2026-08-14). Data as JSON: /api/errors/e442d2e391b75d1e. Report an issue: GitHub.