apache/hadoop · error · IOException

Not enough capacity ratio left on mount: ${mount}, for ${tar

Error message

Not enough capacity ratio left on mount: ${mount}, for ${target}: capacity ratio: ${capacityRatio}. Sum of the capacity ratio of on same disk mount should be <= 1

What it means

Thrown by MountVolumeMap.setCapacityRatio() (MountVolumeMap.java:101) during DataNode startup or volume refresh when same-disk tiering (HDFS-15548) is enabled and the configured capacity ratios of volumes sharing one disk mount sum to more than 1. Ratios come from dfs.datanode.same-disk-tiering.capacity-ratio.percentage entries of the form [ratio]/path (e.g. [0.3]/disk1/archive), and MountVolumeInfo.setCapacityRatio refuses an assignment that overcommits the mount.

Source

Thrown at hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/MountVolumeMap.java:101

  void removeVolume(FsVolumeImpl target) {
    String mount = target.getMount();
    if (!mount.isEmpty()) {
      MountVolumeInfo info = mountVolumeMapping.get(mount);
      info.removeVolume(target);
      if (info.size() == 0) {
        mountVolumeMapping.remove(mount);
      }
    }
  }

  void setCapacityRatio(FsVolumeImpl target, double capacityRatio)
      throws IOException {
    String mount = target.getMount();
    if (!mount.isEmpty()) {
      MountVolumeInfo info = mountVolumeMapping.get(mount);
      if (!info.setCapacityRatio(
          target.getStorageType(), capacityRatio)) {
        throw new IOException(
            "Not enough capacity ratio left on mount: "
                + mount + ", for " + target + ": capacity ratio: "
                + capacityRatio + ". Sum of the capacity"
                + " ratio of on same disk mount should be <= 1");
      }
    }
  }

  public boolean hasMount(String mount) {
    return mountVolumeMapping.containsKey(mount);
  }
}

View on GitHub (pinned to 2add963021)

Solutions

1. Sum the ratios for every configured volume on the same mount and reduce them so the total is <= 1 (e.g. [0.3]/disk1/archive + [0.7]/disk1/disk)
2. Verify each path really lands on the mount you think (df/realpath on the volume dirs) - symlinked paths silently share mounts
3. Keep each individual ratio within 0..1 as StorageLocation.parseCapacityRatio requires
4. If tiering is not intended, remove the capacity-ratio entries or set dfs.datanode.same-disk-tiering.enabled=false and restart the DataNode
5. Apply the corrected value, then re-run refresh or restart; the check re-executes at volume initialization

Example fix

# before: two volumes on /mnt/disk0 overcommit the mount (0.6 + 0.5 > 1)
<property>
  <name>dfs.datanode.same-disk-tiering.capacity-ratio.percentage</name>
  <value>[0.6]/mnt/disk0/archive,[0.5]/mnt/disk0/disk</value>
</property>

# after: ratios on one mount sum to exactly 1
<property>
  <name>dfs.datanode.same-disk-tiering.capacity-ratio.percentage</name>
  <value>[0.3]/mnt/disk0/archive,[0.7]/mnt/disk0/disk</value>
</property>

// before enabling same-disk tiering: parse the config and sum per mount
Map<String, Double> perMount = new HashMap<>();
for (Map.Entry<URI, Double> e : StorageLocation.parseCapacityRatio(conf.get(
        "dfs.datanode.same-disk-tiering.capacity-ratio.percentage", "")).entrySet()) {
  String mount = org.apache.hadoop.fs.FileSystem.getStoreMount(e.getKey()); // or df-based lookup
  double sum = perMount.getOrDefault(mount, 0.0) + e.getValue();
  if (sum > 1.0) {
    throw new IllegalArgumentException("Ratios on mount " + mount + " sum to " + sum);
  }
  perMount.put(mount, sum);
}

When it happens

Trigger: dfs.datanode.same-disk-tiering.enabled=true plus capacity-ratio entries whose ratios on the same mount exceed 1.0 (e.g. [0.6]/mnt/disk/archive and [0.5]/mnt/disk/disk); ratios fine individually but paths resolve to the same mount via symlinks/bind mounts; typo like [1.2]/path passing StorageLocation's 0-1 check is impossible, but two 0.9 entries on one mount is the classic hit.

Common situations: Enabling block tiering (ARCHIVE + DISK volumes on one physical disk) and copy-pasting ratio entries; re-arranging directories so two configured volumes now share a mount; changing one volume's ratio and forgetting the counterpart's; volume refresh (dfsadmin -refreshNodes-style reconfiguration) after editing the ratio property.

Related errors

• Incompatible node types: storageType={storageType} but Stora
• No services to connect, missing NameNode address.
• BlockPoolSliceStorage.recoverTransitionRead: attempt to load
• No block pool offer service for bpid={}
• Not a valid Boolean value for {property} in reconfSlowPeerPa

AI-assisted analysis of apache/hadoop@2add963021 (2026-08-22). Data as JSON: /api/errors/51e919ae4ee49d51. Report an issue: GitHub.