This is an automated email from the ASF dual-hosted git repository.
azagrebin pushed a commit to branch release-1.10
in repository https://gitbox.apache.org/repos/asf/flink.git
commit dfd82556c407c2da51f24f606217a2c9366a6c28
Author: Andrey Zagrebin <azagre...@apache.org>
AuthorDate: Fri Jan 31 13:45:14 2020 +0100
[FLINK-15143][docs] Add tuning and troubleshooting guides for memory
configuration
---
docs/ops/deployment/cluster_setup.md | 2 +-
docs/ops/deployment/cluster_setup.zh.md | 2 +-
docs/ops/memory/mem_setup.md | 4 +-
docs/ops/memory/mem_trouble.md | 74 ++++++++++++++++++++++++++++
docs/ops/memory/mem_trouble.zh.md | 74 ++++++++++++++++++++++++++++
docs/ops/memory/mem_tuning.md | 87 +++++++++++++++++++++++++++++++++
docs/ops/memory/mem_tuning.zh.md | 87 +++++++++++++++++++++++++++++++++
docs/ops/state/state_backends.md | 7 +++
docs/ops/state/state_backends.zh.md | 5 ++
9 files changed, 338 insertions(+), 4 deletions(-)
diff --git a/docs/ops/deployment/cluster_setup.md
b/docs/ops/deployment/cluster_setup.md
index 2fda08c..4f12111 100644
--- a/docs/ops/deployment/cluster_setup.md
+++ b/docs/ops/deployment/cluster_setup.md
@@ -107,7 +107,7 @@ Please see the [configuration page](../config.html) for
details and additional c
In particular,
* the amount of available memory per JobManager (`jobmanager.heap.size`),
- * the amount of available memory per TaskManager
(`taskmanager.memory.process.size`),
+ * the amount of available memory per TaskManager
(`taskmanager.memory.process.size` and check [memory setup
guide](../memory/mem_tuning.html#configure-memory-for-standalone-deployment)),
* the number of available CPUs per machine (`taskmanager.numberOfTaskSlots`),
* the total number of CPUs in the cluster (`parallelism.default`) and
* the temporary directories (`io.tmp.dirs`)
diff --git a/docs/ops/deployment/cluster_setup.zh.md
b/docs/ops/deployment/cluster_setup.zh.md
index 4bd474f..ef8f3e8 100644
--- a/docs/ops/deployment/cluster_setup.zh.md
+++ b/docs/ops/deployment/cluster_setup.zh.md
@@ -107,7 +107,7 @@ Please see the [configuration page](../config.html) for
details and additional c
In particular,
* the amount of available memory per JobManager (`jobmanager.heap.size`),
- * the amount of available memory per TaskManager
(`taskmanager.memory.process.size`),
+ * the amount of available memory per TaskManager
(`taskmanager.memory.process.size` and check [memory setup
guide](../memory/mem_tuning.html#configure-memory-for-standalone-deployment)),
* the number of available CPUs per machine (`taskmanager.numberOfTaskSlots`),
* the total number of CPUs in the cluster (`parallelism.default`) and
* the temporary directories (`io.tmp.dirs`)
diff --git a/docs/ops/memory/mem_setup.md b/docs/ops/memory/mem_setup.md
index 4b2786f..f2a2b06 100644
--- a/docs/ops/memory/mem_setup.md
+++ b/docs/ops/memory/mem_setup.md
@@ -62,7 +62,7 @@ and *direct memory*.
If you configure *total process memory* you declare how much memory in total
should be assigned to the Flink *JVM process*.
For the containerized deployments it corresponds to the size of the requested
container, see also
-[how to configure memory for containers](#heading=h.q0nx4u2c3pzx)
+[how to configure memory for
containers](mem_tuning.html#configure-memory-for-containers)
([Kubernetes](../deployment/kubernetes.html),
[Yarn](../deployment/yarn_setup.html) or [Mesos](../deployment/mesos.html)).
Another way to setup the memory is to set [task
heap](#task-operator-heap-memory) and [managed memory](#managed-memory)
@@ -110,7 +110,7 @@ The size of *managed memory* can be
*Size* will override *fraction*, if both are set.
If neither *size* nor *fraction* is explicitly configured, the [default
fraction](../config.html#taskmanager-memory-managed-fraction) will be used.
-See also [how to configure memory for state backends](#heading=h.srelwz7nbzwa)
and [batch jobs](#heading=h.d6mjc9yd85c0).
+See also [how to configure memory for state
backends](mem_tuning.html#configure-memory-for-state-backends) and [batch
jobs](mem_tuning.html#configure-memory-for-batch-jobs).
## Configure Off-Heap Memory (direct or native)
diff --git a/docs/ops/memory/mem_trouble.md b/docs/ops/memory/mem_trouble.md
new file mode 100644
index 0000000..cd6463b
--- /dev/null
+++ b/docs/ops/memory/mem_trouble.md
@@ -0,0 +1,74 @@
+---
+title: "Troubleshooting"
+nav-parent_id: ops_mem
+nav-pos: 4
+---
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+* toc
+{:toc}
+
+## IllegalConfigurationException
+
+If you see an *IllegalConfigurationException* thrown from
*TaskExecutorProcessUtils*, it usually indicates
+that there is either an invalid configuration value (e.g. negative memory
size, fraction that is greater than 1, etc.)
+or configuration conflicts. Check the documentation chapters related to the
[memory components](mem_setup.html#detailed-memory-model)
+mentioned in the exception message.
+
+## OutOfMemoryError: Java heap space
+
+The exception usually indicates that the JVM heap is too small. You can try to
increase the JVM heap size
+by increasing [total memory](mem_setup.html#configure-total-memory) or [task
heap memory](mem_setup.html#task-operator-heap-memory).
+
+<span class="label label-info">Note</span> You can also increase the
[framework heap memory](mem_detail.html#framework-memory) but this option
+is advanced and should only be changed if you are sure that the Flink
framework itself needs more memory.
+
+## OutOfMemoryError: Direct buffer memory
+
+The exception usually indicates that the JVM *direct memory* limit is too
small or that there is a *direct memory leak*.
+Check whether user code or other external dependencies use the JVM *direct
memory* and that it is properly accounted for.
+You can try to increase its limit by adjusting [direct off-heap
memory](mem_setup.html#detailed-memory-model).
+See also [how to configure off-heap
memory](mem_setup.html#configure-off-heap-memory-direct-or-native) and
+the [JVM arguments](mem_detail.html#jvm-parameters) which Flink sets.
+
+## OutOfMemoryError: Metaspace
+
+The exception usually indicates that [JVM metaspace
limit](mem_detail.html#jvm-parameters) is configured too small.
+You can try to increase the [JVM metaspace
option](../config.html#taskmanager-memory-jvm-metaspace-size).
+
+## IOException: Insufficient number of network buffers
+
+The exception usually indicates that the size of the configured [network
memory](mem_setup.html#detailed-memory-model)
+is not big enough. You can try to increase the *network memory* by adjusting
the following options:
+*
[`taskmanager.memory.network.min`](../config.html#taskmanager-memory-network-min)
+*
[`taskmanager.memory.network.max`](../config.html#taskmanager-memory-network-max)
+*
[`taskmanager.memory.network.fraction`](../config.html#taskmanager-memory-network-fraction)
+
+## Container Memory Exceeded
+
+If a task executor container tries to allocate memory beyond its requested
size (Yarn, Mesos or Kubernetes),
+this usually indicates that Flink has not reserved enough native memory. You
can observe this either by using an external
+monitoring system or from the error messages when a container gets killed by
the deployment environment.
+
+If [RocksDBStateBackend](../state/state_backends.html#the-rocksdbstatebackend)
is used and the memory controlling is disabled,
+you can try to increase the [managed memory](mem_setup.html#managed-memory).
+
+Alternatively, you can increase the [JVM
overhead](mem_setup.html#detailed-memory-model).
+See also [how to configure memory for
containers](mem_tuning.html#configure-memory-for-containers).
diff --git a/docs/ops/memory/mem_trouble.zh.md
b/docs/ops/memory/mem_trouble.zh.md
new file mode 100644
index 0000000..cd6463b
--- /dev/null
+++ b/docs/ops/memory/mem_trouble.zh.md
@@ -0,0 +1,74 @@
+---
+title: "Troubleshooting"
+nav-parent_id: ops_mem
+nav-pos: 4
+---
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+* toc
+{:toc}
+
+## IllegalConfigurationException
+
+If you see an *IllegalConfigurationException* thrown from
*TaskExecutorProcessUtils*, it usually indicates
+that there is either an invalid configuration value (e.g. negative memory
size, fraction that is greater than 1, etc.)
+or configuration conflicts. Check the documentation chapters related to the
[memory components](mem_setup.html#detailed-memory-model)
+mentioned in the exception message.
+
+## OutOfMemoryError: Java heap space
+
+The exception usually indicates that the JVM heap is too small. You can try to
increase the JVM heap size
+by increasing [total memory](mem_setup.html#configure-total-memory) or [task
heap memory](mem_setup.html#task-operator-heap-memory).
+
+<span class="label label-info">Note</span> You can also increase the
[framework heap memory](mem_detail.html#framework-memory) but this option
+is advanced and should only be changed if you are sure that the Flink
framework itself needs more memory.
+
+## OutOfMemoryError: Direct buffer memory
+
+The exception usually indicates that the JVM *direct memory* limit is too
small or that there is a *direct memory leak*.
+Check whether user code or other external dependencies use the JVM *direct
memory* and that it is properly accounted for.
+You can try to increase its limit by adjusting [direct off-heap
memory](mem_setup.html#detailed-memory-model).
+See also [how to configure off-heap
memory](mem_setup.html#configure-off-heap-memory-direct-or-native) and
+the [JVM arguments](mem_detail.html#jvm-parameters) which Flink sets.
+
+## OutOfMemoryError: Metaspace
+
+The exception usually indicates that [JVM metaspace
limit](mem_detail.html#jvm-parameters) is configured too small.
+You can try to increase the [JVM metaspace
option](../config.html#taskmanager-memory-jvm-metaspace-size).
+
+## IOException: Insufficient number of network buffers
+
+The exception usually indicates that the size of the configured [network
memory](mem_setup.html#detailed-memory-model)
+is not big enough. You can try to increase the *network memory* by adjusting
the following options:
+*
[`taskmanager.memory.network.min`](../config.html#taskmanager-memory-network-min)
+*
[`taskmanager.memory.network.max`](../config.html#taskmanager-memory-network-max)
+*
[`taskmanager.memory.network.fraction`](../config.html#taskmanager-memory-network-fraction)
+
+## Container Memory Exceeded
+
+If a task executor container tries to allocate memory beyond its requested
size (Yarn, Mesos or Kubernetes),
+this usually indicates that Flink has not reserved enough native memory. You
can observe this either by using an external
+monitoring system or from the error messages when a container gets killed by
the deployment environment.
+
+If [RocksDBStateBackend](../state/state_backends.html#the-rocksdbstatebackend)
is used and the memory controlling is disabled,
+you can try to increase the [managed memory](mem_setup.html#managed-memory).
+
+Alternatively, you can increase the [JVM
overhead](mem_setup.html#detailed-memory-model).
+See also [how to configure memory for
containers](mem_tuning.html#configure-memory-for-containers).
diff --git a/docs/ops/memory/mem_tuning.md b/docs/ops/memory/mem_tuning.md
new file mode 100644
index 0000000..9d83d00
--- /dev/null
+++ b/docs/ops/memory/mem_tuning.md
@@ -0,0 +1,87 @@
+---
+title: "Memory tuning guide"
+nav-parent_id: ops_mem
+nav-pos: 3
+---
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+In addition to the [main memory setup guide](mem_setup.html), this section
explains how to setup memory of task executors
+depending on the use case and which options are important in which case.
+
+* toc
+{:toc}
+
+## Configure memory for standalone deployment
+
+It is recommended to configure [total Flink
memory](mem_setup.html#configure-total-memory)
+([`taskmanager.memory.flink.size`](../config.html#taskmanager-memory-flink-size))
or its [components](mem_setup.html#detailed-memory-model)
+for [standalone deployment](../deployment/cluster_setup.html) where you want
to declare how much memory is given to Flink itself.
+Additionally, you can adjust *JVM metaspace* if it causes
[problems](mem_trouble.html#outofmemoryerror-metaspace).
+
+The *total Process memory* is not relevant because *JVM overhead* is not
controlled by Flink or deployment environment,
+only physical resources of the executing machine matter in this case.
+
+## Configure memory for containers
+
+It is recommended to configure [total process
memory](mem_setup.html#configure-total-memory)
+([`taskmanager.memory.process.size`](../config.html#taskmanager-memory-process-size))
for the containerized deployments
+([Kubernetes](../deployment/kubernetes.html),
[Yarn](../deployment/yarn_setup.html) or [Mesos](../deployment/mesos.html)).
+It declares how much memory in total should be assigned to the Flink *JVM
process* and corresponds to the size of the requested container.
+
+<span class="label label-info">Note</span> If you configure the *total Flink
memory* Flink will implicitly add JVM memory components
+to derive the *total process memory* and request a container with the memory
of that derived size,
+see also [detailed Memory Model](mem_setup.html#detailed-memory-model).
+
+<div class="alert alert-warning">
+ <strong>Warning:</strong> If Flink or user code allocates unmanaged off-heap
(native) memory beyond the container size
+ the job can fail because the deployment environment can kill the offending
containers.
+</div>
+See also description of [container memory
exceeded](mem_trouble.html#container-memory-exceeded) failure.
+
+## Configure memory for state backends
+
+When deploying a Flink streaming application, the type of [state
backend](../state/state_backends.html) used
+will dictate the optimal memory configurations of your cluster.
+
+### Heap state backend
+
+When running a stateless job or using a heap state backend
([MemoryStateBackend](../state/state_backends.html#the-memorystatebackend)
+or [FsStateBackend](../state/state_backends.html#the-fsstatebackend), set
[managed memory](mem_setup.html#managed-memory) to zero.
+This will ensure that the maximum amount of memory is allocated for user code
on the JVM.
+
+### RocksDB state backend
+
+The
[RocksDBStateBackend](../state/state_backends.html#the-rocksdbstatebackend)
uses native memory. By default,
+RocksDB is setup to limit native memory allocation to the size of the [managed
memory](mem_setup.html#managed-memory).
+Therefore, it is important to reserve enough *managed memory* for your state
use case. If you disable the default RocksDB memory control,
+task executors can be killed in containerized deployments if RocksDB allocates
memory above the limit of the requested container size
+(the [total process memory](mem_setup.html#configure-total-memory)).
+See also [how to tune RocksDB
memory](../state/large_state_tuning.html#tuning-rocksdb-memory)
+and
[state.backend.rocksdb.memory.managed](../config.html#state-backend-rocksdb-memory-managed).
+
+## Configure memory for batch jobs
+
+Flink's batch operators leverage [managed
memory](../memory/mem_setup.html#managed-memory) to run more efficiently.
+In doing so, some operations can be performed directly on raw data without
having to be deserialized into Java objects.
+This means that [managed memory](../memory/mem_setup.html#managed-memory)
configurations have practical effects
+on the performance of your applications. Flink will attempt to allocate and
use as much [managed memory](../memory/mem_setup.html#managed-memory)
+as configured for batch jobs but not go beyond its limits. This prevents
`OutOfMemoryError`'s because Flink knows precisely
+how much memory it has to leverage. If the [managed
memory](../memory/mem_setup.html#managed-memory) is not sufficient,
+Flink will gracefully spill to disk.
diff --git a/docs/ops/memory/mem_tuning.zh.md b/docs/ops/memory/mem_tuning.zh.md
new file mode 100644
index 0000000..9d83d00
--- /dev/null
+++ b/docs/ops/memory/mem_tuning.zh.md
@@ -0,0 +1,87 @@
+---
+title: "Memory tuning guide"
+nav-parent_id: ops_mem
+nav-pos: 3
+---
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+In addition to the [main memory setup guide](mem_setup.html), this section
explains how to setup memory of task executors
+depending on the use case and which options are important in which case.
+
+* toc
+{:toc}
+
+## Configure memory for standalone deployment
+
+It is recommended to configure [total Flink
memory](mem_setup.html#configure-total-memory)
+([`taskmanager.memory.flink.size`](../config.html#taskmanager-memory-flink-size))
or its [components](mem_setup.html#detailed-memory-model)
+for [standalone deployment](../deployment/cluster_setup.html) where you want
to declare how much memory is given to Flink itself.
+Additionally, you can adjust *JVM metaspace* if it causes
[problems](mem_trouble.html#outofmemoryerror-metaspace).
+
+The *total Process memory* is not relevant because *JVM overhead* is not
controlled by Flink or deployment environment,
+only physical resources of the executing machine matter in this case.
+
+## Configure memory for containers
+
+It is recommended to configure [total process
memory](mem_setup.html#configure-total-memory)
+([`taskmanager.memory.process.size`](../config.html#taskmanager-memory-process-size))
for the containerized deployments
+([Kubernetes](../deployment/kubernetes.html),
[Yarn](../deployment/yarn_setup.html) or [Mesos](../deployment/mesos.html)).
+It declares how much memory in total should be assigned to the Flink *JVM
process* and corresponds to the size of the requested container.
+
+<span class="label label-info">Note</span> If you configure the *total Flink
memory* Flink will implicitly add JVM memory components
+to derive the *total process memory* and request a container with the memory
of that derived size,
+see also [detailed Memory Model](mem_setup.html#detailed-memory-model).
+
+<div class="alert alert-warning">
+ <strong>Warning:</strong> If Flink or user code allocates unmanaged off-heap
(native) memory beyond the container size
+ the job can fail because the deployment environment can kill the offending
containers.
+</div>
+See also description of [container memory
exceeded](mem_trouble.html#container-memory-exceeded) failure.
+
+## Configure memory for state backends
+
+When deploying a Flink streaming application, the type of [state
backend](../state/state_backends.html) used
+will dictate the optimal memory configurations of your cluster.
+
+### Heap state backend
+
+When running a stateless job or using a heap state backend
([MemoryStateBackend](../state/state_backends.html#the-memorystatebackend)
+or [FsStateBackend](../state/state_backends.html#the-fsstatebackend), set
[managed memory](mem_setup.html#managed-memory) to zero.
+This will ensure that the maximum amount of memory is allocated for user code
on the JVM.
+
+### RocksDB state backend
+
+The
[RocksDBStateBackend](../state/state_backends.html#the-rocksdbstatebackend)
uses native memory. By default,
+RocksDB is setup to limit native memory allocation to the size of the [managed
memory](mem_setup.html#managed-memory).
+Therefore, it is important to reserve enough *managed memory* for your state
use case. If you disable the default RocksDB memory control,
+task executors can be killed in containerized deployments if RocksDB allocates
memory above the limit of the requested container size
+(the [total process memory](mem_setup.html#configure-total-memory)).
+See also [how to tune RocksDB
memory](../state/large_state_tuning.html#tuning-rocksdb-memory)
+and
[state.backend.rocksdb.memory.managed](../config.html#state-backend-rocksdb-memory-managed).
+
+## Configure memory for batch jobs
+
+Flink's batch operators leverage [managed
memory](../memory/mem_setup.html#managed-memory) to run more efficiently.
+In doing so, some operations can be performed directly on raw data without
having to be deserialized into Java objects.
+This means that [managed memory](../memory/mem_setup.html#managed-memory)
configurations have practical effects
+on the performance of your applications. Flink will attempt to allocate and
use as much [managed memory](../memory/mem_setup.html#managed-memory)
+as configured for batch jobs but not go beyond its limits. This prevents
`OutOfMemoryError`'s because Flink knows precisely
+how much memory it has to leverage. If the [managed
memory](../memory/mem_setup.html#managed-memory) is not sufficient,
+Flink will gracefully spill to disk.
diff --git a/docs/ops/state/state_backends.md b/docs/ops/state/state_backends.md
index 5bc0afb..134621b 100644
--- a/docs/ops/state/state_backends.md
+++ b/docs/ops/state/state_backends.md
@@ -74,6 +74,8 @@ The MemoryStateBackend is encouraged for:
- Local development and debugging
- Jobs that do hold little state, such as jobs that consist only of
record-at-a-time functions (Map, FlatMap, Filter, ...). The Kafka Consumer
requires very little state.
+It is also recommended to set [managed
memory](../memory/mem_setup.html#managed-memory) to zero.
+This will ensure that the maximum amount of memory is allocated for user code
on the JVM.
### The FsStateBackend
@@ -92,6 +94,9 @@ The FsStateBackend is encouraged for:
- Jobs with large state, long windows, large key/value states.
- All high-availability setups.
+It is also recommended to set [managed
memory](../memory/mem_setup.html#managed-memory) to zero.
+This will ensure that the maximum amount of memory is allocated for user code
on the JVM.
+
### The RocksDBStateBackend
The *RocksDBStateBackend* is configured with a file system URL (type, address,
path), such as "hdfs://namenode:40010/flink/checkpoints" or
"file:///data/flink/checkpoints".
@@ -119,6 +124,8 @@ This also means, however, that the maximum throughput that
can be achieved will
this state backend. All reads/writes from/to this backend have to go through
de-/serialization to retrieve/store the state objects, which is also more
expensive than always working with the
on-heap representation as the heap-based backends are doing.
+Check also recommendations about the [task executor memory
configuration](../memory/mem_tuning.html#rocksdb-state-backend) for the
RocksDBStateBackend.
+
RocksDBStateBackend is currently the only backend that offers incremental
checkpoints (see [here](large_state_tuning.html)).
Certain RocksDB native metrics are available but disabled by default, you can
find full documentation [here]({{ site.baseurl
}}/ops/config.html#rocksdb-native-metrics)
diff --git a/docs/ops/state/state_backends.zh.md
b/docs/ops/state/state_backends.zh.md
index 60718c2..57feabd 100644
--- a/docs/ops/state/state_backends.zh.md
+++ b/docs/ops/state/state_backends.zh.md
@@ -71,6 +71,7 @@ MemoryStateBackend 适用场景：
- 本地开发和调试。
- 状态很小的 Job，例如：由每次只处理一条记录的函数（Map、FlatMap、Filter 等）构成的 Job。Kafka Consumer
仅仅需要非常小的状态。
+建议同时将 [managed memory](../memory/mem_setup.html#managed-memory)
设为0，以保证将最大限度的内存分配给 JVM 上的用户代码。
### FsStateBackend
@@ -91,6 +92,8 @@ FsStateBackend 适用场景:
- 状态比较大、窗口比较长、key/value 状态比较大的 Job。
- 所有高可用的场景。
+建议同时将 [managed memory](../memory/mem_setup.html#managed-memory)
设为0，以保证将最大限度的内存分配给 JVM 上的用户代码。
+
### RocksDBStateBackend
*RocksDBStateBackend* 需要配置一个文件系统的 URL
（类型、地址、路径），例如："hdfs://namenode:40010/flink/checkpoints" 或
"file:///data/flink/checkpoints"。
@@ -115,6 +118,8 @@ RocksDBStateBackend 的适用场景：
然而，这也意味着使用 RocksDBStateBackend 将会使应用程序的最大吞吐量降低。
所有的读写都必须序列化、反序列化操作，这个比基于堆内存的 state backend 的效率要低很多。
+请同时参考 [Task Executor 内存配置](../memory/mem_tuning.html#rocksdb-state-backend)
中关于 RocksDBStateBackend 的建议。
+
RocksDBStateBackend 是目前唯一支持增量 CheckPoint 的 State Backend (见
[这里](large_state_tuning.html))。
可以使用一些 RocksDB 的本地指标(metrics)，但默认是关闭的。你能在 [这里]({{ site.baseurl
}}/zh/ops/config.html#rocksdb-native-metrics) 找到关于 RocksDB 本地指标的文档。
