YARN-4330. MiniYARNCluster is showing multiple Failed to instantiate default resource calculator warning messages. Contributed by Varun Saxena
This commit is contained in:
parent
4667564162
commit
3541ed8068
@ -190,6 +190,9 @@ public static ResourceCalculatorPlugin getResourceCalculatorPlugin(
|
||||
}
|
||||
try {
|
||||
return new ResourceCalculatorPlugin();
|
||||
} catch (UnsupportedOperationException ue) {
|
||||
LOG.warn("Failed to instantiate default resource calculator. "
|
||||
+ ue.getMessage());
|
||||
} catch (Throwable t) {
|
||||
LOG.warn(t + ": Failed to instantiate default resource calculator.", t);
|
||||
}
|
||||
|
@ -1344,7 +1344,8 @@
|
||||
</property>
|
||||
|
||||
<property>
|
||||
<description>How often to monitor the node and the containers.</description>
|
||||
<description>How often to monitor the node and the containers.
|
||||
If 0 or negative, monitoring is disabled.</description>
|
||||
<name>yarn.nodemanager.resource-monitor.interval-ms</name>
|
||||
<value>3000</value>
|
||||
</property>
|
||||
@ -1362,7 +1363,8 @@
|
||||
|
||||
<property>
|
||||
<description>How often to monitor containers. If not set, the value for
|
||||
yarn.nodemanager.resource-monitor.interval-ms will be used.</description>
|
||||
yarn.nodemanager.resource-monitor.interval-ms will be used.
|
||||
If 0 or negative, container monitoring is disabled.</description>
|
||||
<name>yarn.nodemanager.container-monitor.interval-ms</name>
|
||||
</property>
|
||||
|
||||
|
@ -78,6 +78,11 @@ protected void serviceInit(Configuration conf) throws Exception {
|
||||
* @return <em>true</em> if we can monitor the node resource utilization.
|
||||
*/
|
||||
private boolean isEnabled() {
|
||||
if (this.monitoringInterval <= 0) {
|
||||
LOG.info("Node Resource monitoring interval is <=0. "
|
||||
+ this.getClass().getName() + " is disabled.");
|
||||
return false;
|
||||
}
|
||||
if (resourceCalculatorPlugin == null) {
|
||||
LOG.info("ResourceCalculatorPlugin is unavailable on this system. "
|
||||
+ this.getClass().getName() + " is disabled.");
|
||||
|
@ -172,7 +172,8 @@ protected void serviceInit(Configuration myConf) throws Exception {
|
||||
LOG.info("Physical memory check enabled: " + pmemCheckEnabled);
|
||||
LOG.info("Virtual memory check enabled: " + vmemCheckEnabled);
|
||||
|
||||
containersMonitorEnabled = isContainerMonitorEnabled();
|
||||
containersMonitorEnabled =
|
||||
isContainerMonitorEnabled() && monitoringInterval > 0;
|
||||
LOG.info("ContainersMonitor enabled: " + containersMonitorEnabled);
|
||||
|
||||
nodeCpuPercentageForYARN =
|
||||
|
@ -37,6 +37,12 @@ public class NodeManagerHardwareUtils {
|
||||
private static final Log LOG = LogFactory
|
||||
.getLog(NodeManagerHardwareUtils.class);
|
||||
|
||||
private static boolean isHardwareDetectionEnabled(Configuration conf) {
|
||||
return conf.getBoolean(
|
||||
YarnConfiguration.NM_ENABLE_HARDWARE_CAPABILITY_DETECTION,
|
||||
YarnConfiguration.DEFAULT_NM_ENABLE_HARDWARE_CAPABILITY_DETECTION);
|
||||
}
|
||||
|
||||
/**
|
||||
*
|
||||
* Returns the number of CPUs on the node. This value depends on the
|
||||
@ -138,6 +144,15 @@ public static int getNodeCpuPercentage(Configuration conf) {
|
||||
return nodeCpuPercentage;
|
||||
}
|
||||
|
||||
private static int getConfiguredVCores(Configuration conf) {
|
||||
int cores = conf.getInt(YarnConfiguration.NM_VCORES,
|
||||
YarnConfiguration.DEFAULT_NM_VCORES);
|
||||
if (cores == -1) {
|
||||
cores = YarnConfiguration.DEFAULT_NM_VCORES;
|
||||
}
|
||||
return cores;
|
||||
}
|
||||
|
||||
/**
|
||||
* Function to return the number of vcores on the system that can be used for
|
||||
* YARN containers. If a number is specified in the configuration file, then
|
||||
@ -154,11 +169,16 @@ public static int getNodeCpuPercentage(Configuration conf) {
|
||||
*
|
||||
*/
|
||||
public static int getVCores(Configuration conf) {
|
||||
if (!isHardwareDetectionEnabled(conf)) {
|
||||
return getConfiguredVCores(conf);
|
||||
}
|
||||
// is this os for which we can determine cores?
|
||||
ResourceCalculatorPlugin plugin =
|
||||
ResourceCalculatorPlugin.getResourceCalculatorPlugin(null, conf);
|
||||
|
||||
return NodeManagerHardwareUtils.getVCores(plugin, conf);
|
||||
if (plugin == null) {
|
||||
return getConfiguredVCores(conf);
|
||||
}
|
||||
return getVCoresInternal(plugin, conf);
|
||||
}
|
||||
|
||||
/**
|
||||
@ -180,43 +200,35 @@ public static int getVCores(Configuration conf) {
|
||||
*/
|
||||
public static int getVCores(ResourceCalculatorPlugin plugin,
|
||||
Configuration conf) {
|
||||
if (!isHardwareDetectionEnabled(conf) || plugin == null) {
|
||||
return getConfiguredVCores(conf);
|
||||
}
|
||||
return getVCoresInternal(plugin, conf);
|
||||
}
|
||||
|
||||
int cores;
|
||||
boolean hardwareDetectionEnabled =
|
||||
conf.getBoolean(
|
||||
YarnConfiguration.NM_ENABLE_HARDWARE_CAPABILITY_DETECTION,
|
||||
YarnConfiguration.DEFAULT_NM_ENABLE_HARDWARE_CAPABILITY_DETECTION);
|
||||
|
||||
private static int getVCoresInternal(ResourceCalculatorPlugin plugin,
|
||||
Configuration conf) {
|
||||
String message;
|
||||
if (!hardwareDetectionEnabled || plugin == null) {
|
||||
cores =
|
||||
conf.getInt(YarnConfiguration.NM_VCORES,
|
||||
YarnConfiguration.DEFAULT_NM_VCORES);
|
||||
if (cores == -1) {
|
||||
cores = YarnConfiguration.DEFAULT_NM_VCORES;
|
||||
}
|
||||
} else {
|
||||
cores = conf.getInt(YarnConfiguration.NM_VCORES, -1);
|
||||
if (cores == -1) {
|
||||
float physicalCores =
|
||||
NodeManagerHardwareUtils.getContainersCPUs(plugin, conf);
|
||||
float multiplier =
|
||||
conf.getFloat(YarnConfiguration.NM_PCORES_VCORES_MULTIPLIER,
|
||||
YarnConfiguration.DEFAULT_NM_PCORES_VCORES_MULTIPLIER);
|
||||
if (multiplier > 0) {
|
||||
float tmp = physicalCores * multiplier;
|
||||
if (tmp > 0 && tmp < 1) {
|
||||
// on a single core machine - tmp can be between 0 and 1
|
||||
cores = 1;
|
||||
} else {
|
||||
cores = (int) tmp;
|
||||
}
|
||||
int cores = conf.getInt(YarnConfiguration.NM_VCORES, -1);
|
||||
if (cores == -1) {
|
||||
float physicalCores =
|
||||
NodeManagerHardwareUtils.getContainersCPUs(plugin, conf);
|
||||
float multiplier =
|
||||
conf.getFloat(YarnConfiguration.NM_PCORES_VCORES_MULTIPLIER,
|
||||
YarnConfiguration.DEFAULT_NM_PCORES_VCORES_MULTIPLIER);
|
||||
if (multiplier > 0) {
|
||||
float tmp = physicalCores * multiplier;
|
||||
if (tmp > 0 && tmp < 1) {
|
||||
// on a single core machine - tmp can be between 0 and 1
|
||||
cores = 1;
|
||||
} else {
|
||||
message = "Illegal value for "
|
||||
+ YarnConfiguration.NM_PCORES_VCORES_MULTIPLIER
|
||||
+ ". Value must be greater than 0.";
|
||||
throw new IllegalArgumentException(message);
|
||||
cores = (int) tmp;
|
||||
}
|
||||
} else {
|
||||
message = "Illegal value for "
|
||||
+ YarnConfiguration.NM_PCORES_VCORES_MULTIPLIER
|
||||
+ ". Value must be greater than 0.";
|
||||
throw new IllegalArgumentException(message);
|
||||
}
|
||||
}
|
||||
if(cores <= 0) {
|
||||
@ -228,6 +240,15 @@ public static int getVCores(ResourceCalculatorPlugin plugin,
|
||||
return cores;
|
||||
}
|
||||
|
||||
private static int getConfiguredMemoryMB(Configuration conf) {
|
||||
int memoryMb = conf.getInt(YarnConfiguration.NM_PMEM_MB,
|
||||
YarnConfiguration.DEFAULT_NM_PMEM_MB);
|
||||
if (memoryMb == -1) {
|
||||
memoryMb = YarnConfiguration.DEFAULT_NM_PMEM_MB;
|
||||
}
|
||||
return memoryMb;
|
||||
}
|
||||
|
||||
/**
|
||||
* Function to return how much memory we should set aside for YARN containers.
|
||||
* If a number is specified in the configuration file, then that number is
|
||||
@ -244,8 +265,15 @@ public static int getVCores(ResourceCalculatorPlugin plugin,
|
||||
* @return the amount of memory that will be used for YARN containers in MB.
|
||||
*/
|
||||
public static int getContainerMemoryMB(Configuration conf) {
|
||||
return NodeManagerHardwareUtils.getContainerMemoryMB(
|
||||
ResourceCalculatorPlugin.getResourceCalculatorPlugin(null, conf), conf);
|
||||
if (!isHardwareDetectionEnabled(conf)) {
|
||||
return getConfiguredMemoryMB(conf);
|
||||
}
|
||||
ResourceCalculatorPlugin plugin =
|
||||
ResourceCalculatorPlugin.getResourceCalculatorPlugin(null, conf);
|
||||
if (plugin == null) {
|
||||
return getConfiguredMemoryMB(conf);
|
||||
}
|
||||
return getContainerMemoryMBInternal(plugin, conf);
|
||||
}
|
||||
|
||||
/**
|
||||
@ -267,41 +295,35 @@ public static int getContainerMemoryMB(Configuration conf) {
|
||||
*/
|
||||
public static int getContainerMemoryMB(ResourceCalculatorPlugin plugin,
|
||||
Configuration conf) {
|
||||
if (!isHardwareDetectionEnabled(conf) || plugin == null) {
|
||||
return getConfiguredMemoryMB(conf);
|
||||
}
|
||||
return getContainerMemoryMBInternal(plugin, conf);
|
||||
}
|
||||
|
||||
int memoryMb;
|
||||
boolean hardwareDetectionEnabled = conf.getBoolean(
|
||||
YarnConfiguration.NM_ENABLE_HARDWARE_CAPABILITY_DETECTION,
|
||||
YarnConfiguration.DEFAULT_NM_ENABLE_HARDWARE_CAPABILITY_DETECTION);
|
||||
|
||||
if (!hardwareDetectionEnabled || plugin == null) {
|
||||
memoryMb = conf.getInt(YarnConfiguration.NM_PMEM_MB,
|
||||
YarnConfiguration.DEFAULT_NM_PMEM_MB);
|
||||
if (memoryMb == -1) {
|
||||
memoryMb = YarnConfiguration.DEFAULT_NM_PMEM_MB;
|
||||
private static int getContainerMemoryMBInternal(ResourceCalculatorPlugin plugin,
|
||||
Configuration conf) {
|
||||
int memoryMb = conf.getInt(YarnConfiguration.NM_PMEM_MB, -1);
|
||||
if (memoryMb == -1) {
|
||||
int physicalMemoryMB =
|
||||
(int) (plugin.getPhysicalMemorySize() / (1024 * 1024));
|
||||
int hadoopHeapSizeMB =
|
||||
(int) (Runtime.getRuntime().maxMemory() / (1024 * 1024));
|
||||
int containerPhysicalMemoryMB =
|
||||
(int) (0.8f * (physicalMemoryMB - (2 * hadoopHeapSizeMB)));
|
||||
int reservedMemoryMB =
|
||||
conf.getInt(YarnConfiguration.NM_SYSTEM_RESERVED_PMEM_MB, -1);
|
||||
if (reservedMemoryMB != -1) {
|
||||
containerPhysicalMemoryMB = physicalMemoryMB - reservedMemoryMB;
|
||||
}
|
||||
} else {
|
||||
memoryMb = conf.getInt(YarnConfiguration.NM_PMEM_MB, -1);
|
||||
if (memoryMb == -1) {
|
||||
int physicalMemoryMB =
|
||||
(int) (plugin.getPhysicalMemorySize() / (1024 * 1024));
|
||||
int hadoopHeapSizeMB =
|
||||
(int) (Runtime.getRuntime().maxMemory() / (1024 * 1024));
|
||||
int containerPhysicalMemoryMB =
|
||||
(int) (0.8f * (physicalMemoryMB - (2 * hadoopHeapSizeMB)));
|
||||
int reservedMemoryMB =
|
||||
conf.getInt(YarnConfiguration.NM_SYSTEM_RESERVED_PMEM_MB, -1);
|
||||
if (reservedMemoryMB != -1) {
|
||||
containerPhysicalMemoryMB = physicalMemoryMB - reservedMemoryMB;
|
||||
}
|
||||
if(containerPhysicalMemoryMB <= 0) {
|
||||
LOG.error("Calculated memory for YARN containers is too low."
|
||||
+ " Node memory is " + physicalMemoryMB
|
||||
+ " MB, system reserved memory is "
|
||||
+ reservedMemoryMB + " MB.");
|
||||
}
|
||||
containerPhysicalMemoryMB = Math.max(containerPhysicalMemoryMB, 0);
|
||||
memoryMb = containerPhysicalMemoryMB;
|
||||
if(containerPhysicalMemoryMB <= 0) {
|
||||
LOG.error("Calculated memory for YARN containers is too low."
|
||||
+ " Node memory is " + physicalMemoryMB
|
||||
+ " MB, system reserved memory is "
|
||||
+ reservedMemoryMB + " MB.");
|
||||
}
|
||||
containerPhysicalMemoryMB = Math.max(containerPhysicalMemoryMB, 0);
|
||||
memoryMb = containerPhysicalMemoryMB;
|
||||
}
|
||||
if(memoryMb <= 0) {
|
||||
String message = "Illegal value for " + YarnConfiguration.NM_PMEM_MB
|
||||
|
@ -552,13 +552,16 @@ protected synchronized void serviceInit(Configuration conf)
|
||||
.setNMWebAppHostNameAndPort(config,
|
||||
MiniYARNCluster.getHostname(), 0);
|
||||
|
||||
config.setBoolean(
|
||||
YarnConfiguration.NM_ENABLE_HARDWARE_CAPABILITY_DETECTION, false);
|
||||
// Disable resource checks by default
|
||||
if (!config.getBoolean(
|
||||
YarnConfiguration.YARN_MINICLUSTER_CONTROL_RESOURCE_MONITORING,
|
||||
YarnConfiguration.
|
||||
DEFAULT_YARN_MINICLUSTER_CONTROL_RESOURCE_MONITORING)) {
|
||||
config.setBoolean(YarnConfiguration.NM_PMEM_CHECK_ENABLED, false);
|
||||
config.setBoolean(YarnConfiguration.NM_VMEM_CHECK_ENABLED, false);
|
||||
config.setBoolean(
|
||||
YarnConfiguration.NM_CONTAINER_MONITOR_ENABLED, false);
|
||||
config.setLong(YarnConfiguration.NM_RESOURCE_MON_INTERVAL_MS, 0);
|
||||
}
|
||||
|
||||
LOG.info("Starting NM: " + index);
|
||||
|
Loading…
Reference in New Issue
Block a user