Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Improve the documentation for the Jetty Server Dump #12205

Open
wants to merge 7 commits into
base: jetty-12.0.x
Choose a base branch
from
Next Next commit
Improvements to the jetty server dump documentation
Signed-off-by: Lachlan Roberts <[email protected]>
  • Loading branch information
lachlan-roberts committed Aug 28, 2024
commit 06746304b42d9d911d049fc0daabbdaeefc77be8
Original file line number Diff line number Diff line change
Expand Up @@ -281,4 +281,13 @@ public void beanRemoved(Container parent, Object child)
parent.addBean(younger);
// end::containerSiblings[]
}

public void dumpExample()
{
Server server = new Server();
// tag::dumpExample[]
server.setDumpAfterStart(true);
server.setDumpBeforeStop(true);
// end::dumpExample[]
}
}
Original file line number Diff line number Diff line change
Expand Up @@ -15,101 +15,15 @@

To troubleshoot Jetty when used as a standalone server, there are two main tools: the Jetty Server Dump and enabling DEBUG level logging.

Jetty is based on components organized as a tree, with the `Server` instance at the root of the tree.

As explained in the xref:jmx/index.adoc[JMX section], these components can be exported as JMX MBeans and therefore be accessible from JMX Consoles such as Java Missions Control (JMC).

Being able to take a snapshot of the state of Jetty while it is running is the most useful information that can be attached when reporting an issue.
Such state includes:

* The thread pool configuration and its current state, including how many threads are in use, and their stack trace.
* The TLS configuration.
* The I/O configuration and its current state, including the ports Jetty listens to, how many connections are currently open, and he state of each connection, and the state of the request/response handling for each connection.
* The `Handler` structure and its configuration.
* The web applications deployed and their configurations, including the class loader information.

The prerequisite for troubleshooting is to enable JMX, so that Jetty -- possibly a production server -- can be accessed from a remote location to obtain the information exported via JMX, and possibly be able to reconfigure Jetty to solve the issue.
A useful tool for troubleshooting is JMX, so that Jetty -- possibly a production server -- can be accessed from a remote location to obtain the information exported via JMX, and possibly be able to reconfigure Jetty to solve the issue.

IMPORTANT: Make sure you read about how to secure the access to Jetty when using xref:jmx/index.adoc#remote[remote JMX].

[[dump]]
== Server Dump

The Jetty Server Dump is obtained by invoking, via JMX, the `Server.dump()` operation, as shown below using https://adoptium.net/jmc.html[Java Mission Control (JMC)]:

image::jmc-server-dump.png[]

Find the `Server` MBean in the MBean Tree, under `org.eclipse.jetty.server:type=server,id=0`.
Then click on the "Operations" tab, select the `dump()` operation, and then click the `Execute` button.
In the bottom panel you will see the result of the invocation, that you can copy into a text editor and save to your file system.

[CAUTION]
====
Taking a Jetty Server Dump is a relatively expensive operation, as it dumps the state of all connections (which can be thousands), and the state of all threads.

The result of the invocation may produce a large string, possibly few MiB, that may impact the server memory usage.

Furthermore, dumping the state of the I/O Jetty components takes a little CPU time off the handling of the actual I/O, possibly slowing it down temporarily.

While the slow-down caused by taking the Jetty Server Dump may be noticeable on highly loaded systems, it is typically a very small price to pay to obtain the information about the Jetty state that may be critical to the resolution of an issue.
====

[NOTE]
====
The format of the Jetty Server Dump output is subject to change at any time, as Jetty developers modify the Jetty code and decide to include more state, or remove state that is no longer relevant.

The Jetty Server Dump is organized in a tree whose structure is similar to the runtime Jetty component tree.

At the end of the dump output there is a legend that explains the type of tree node: whether it is a node that represent a _managed_ component, or an _array_ node (or a _map_ node) that represent some component state, etc.
====

[[dump-start-stop]]
=== Dump at Server Start/Stop

The `Server.dump()` operation may also be invoked just after the `Server` starts (to log the state of the freshly started server), and just before the `Server` stops (which may be useful to log the state of server that is not working properly).

You can temporarily enable the Jetty Server Dump at start time by overriding the `jetty.server.dumpAfterStart` property on the command line:

[source,subs=+quotes]
----
$ java -jar $JETTY_HOME/start.jar *jetty.server.dumpAfterStart=true*
----

To make this change persistent across server restarts, see the xref:modules/standard.adoc#server[`server` module] configuration for more information about how to configure the server to dump at start/stop time.

[[dump-detailed]]
=== Detailed ThreadPool Information

By default, the dump of the thread pool will only dump the topmost stack frame of each thread.
It is possible to configure the thread pool to dump the whole stack trace for each thread; while this may be a little more expensive, it provides complete information about the state of each thread, which may be important to diagnose the issue.

See the xref:modules/standard.adoc#threadpool[`threadpool` module] configuration for more information about how to configure the thread pool to dump detailed thread information.

Detailed thread pool information can also be turned on/off on-the-fly via JMX, by finding the `ThreadPool` MBean under `org.eclipse.jetty.util.thread:type=queuedthreadpool,id=0`, then selecting the `detailedDump` attribute and setting it to `true`. You can now perform the `Server.dump()` operation as explained above, and then set `detailedDump` back to `false`.

[[dump-example]]
=== Dump Example

Below you can find a simple example of a Jetty Server Dump, with annotations for the principal components:

[jetty.small%nowrap]
....
[jetty]
setupArgs=--add-modules=http
args=jetty.http.selectors=1 jetty.http.acceptors=1 jetty.threadPool.minThreads=4 jetty.server.dumpAfterStart=true
delete=^[0-9]\{4}
callouts= <$N>,Server@,= QueuedThreadPool,HandlerCollection@,= ServerConnector,ManagedSelector@,keys @,startJarLoader@,unmanaged
....
ifdef::run-jetty[]
<1> The `Server` instance at the root of the tree
<2> The thread pool component
<3> The root of the `Handler` structure
<4> The connector listening on port `8080` for the HTTP/1.1 protocol
<5> A selector component that manages connections
<6> The connections currently managed by the selector component
<7> The server `ClassLoader` and its classpath
<8> The legend for the dump nodes
endif::[]
The Jetty Server Dump is a useful tool for debugging the state of the server.
See the xref:programming-guide:troubleshooting/component-dump.adoc[Jetty Server Dump] section for detailed information.

[[logging]]
== Enabling DEBUG Logging
Expand All @@ -125,7 +39,7 @@ Enabling DEBUG level logging for `org.eclipse.jetty` is very, *very* expensive.
Your server could be slowed down to almost a halt, especially if it is under heavy load.
Furthermore, the log file could quickly fill up the entire filesystem (unless configured to roll over), so you want to be really careful using DEBUG logging.

For production servers, consider using the <<dump,Jetty Server Dump>> first, and enable DEBUG logging only as a last resort.
For production servers, consider using the xref:programming-guide:troubleshooting/component-dump.adoc[Jetty Server Dump] first, and enable DEBUG logging only as a last resort.
====

However, sometimes issues are such that only DEBUG logging can really tell what's going on in the system, and enabling DEBUG logging is your best chance to figure the issue out.
Expand Down
Original file line number Diff line number Diff line change
@@ -1,26 +1,103 @@
//
// ========================================================================
// Copyright (c) 1995 Mort Bay Consulting Pty Ltd and others.
//
// This program and the accompanying materials are made available under the
// terms of the Eclipse Public License v. 2.0 which is available at
// https://www.eclipse.org/legal/epl-2.0, or the Apache License, Version 2.0
// which is available at https://www.apache.org/licenses/LICENSE-2.0.
//
// SPDX-License-Identifier: EPL-2.0 OR Apache-2.0
// ========================================================================
//
[[dump]]
= Jetty Server Dump

= Component Tree Dump
Jetty is based on components organized as a tree, with the `Server` instance at the root of the tree.

Jetty components are organized in a xref:arch/bean.adoc[component tree].
As explained in the xref:operations-guide:jmx/index.adoc[JMX section], these components can be exported as JMX MBeans and therefore be accessible from JMX Consoles such as Java Missions Control (JMC).

At the root of the component tree there is typically a `ContainerLifeCycle` instance -- typically a `Server` instance on the server and an `HttpClient` instance on the client.
Being able to take a snapshot of the state of Jetty while it is running is the most useful information that can be attached when reporting an issue.
Such state includes:

`ContainerLifeCycle` has built-in _dump_ APIs that can be invoked either directly on the `Server` instance, or xref:arch/jmx.adoc[via JMX].
* The thread pool configuration and its current state, including how many threads are in use, and their stack trace.
* The TLS configuration.
* The I/O configuration and its current state, including the ports Jetty listens to, how many connections are currently open, and he state of each connection, and the state of the request/response handling for each connection.
* The `Handler` structure and its configuration.
* The web applications deployed and their configurations, including the class loader information.

You can invoke `Server.dump()` via JMX using a JMX console such as https://adoptium.net/jmc.html[Java Mission Control (JMC)]:
[[dump-start-stop]]
== Dump at Server Start/Stop

The `Server.dump()` operation may also be invoked just after the `Server` starts (to log the state of the freshly started server), and just before the `Server` stops (which may be useful to log the state of server that is not working properly).

You can temporarily enable the Jetty Server Dump at start time by overriding the `jetty.server.dumpAfterStart` property on the command line:

For embedded usage this can be used by calling the setters directly.
[,java,indent=0]
----
include::code:example$src/main/java/org/eclipse/jetty/docs/programming/ComponentDocs.java[tags=dumpExample]
----

Standalone Jetty uses two properties to control this behaviour which are referenced in `jetty.xml` to call these setters.
These properties are `jetty.server.dumpAfterStart` and `jetty.server.dumpBeforeStop`.
lachlan-roberts marked this conversation as resolved.
Show resolved Hide resolved

These can be temporarily enabled by supplying these properties as command line arguments,
or to make this change persistent across server restarts they can be enabled via the `server.ini` file (see xref:operations-guide:modules/standard.adoc#server-config[Server Configuration]).

[source,subs=+quotes]
----
$ java -jar $JETTY_HOME/start.jar *jetty.server.dumpAfterStart=true*
----

== Obtaining Dump via JMX

The Jetty Server Dump can be obtained by invoking, via JMX, the `Server.dump()` operation, as shown below using https://adoptium.net/jmc.html[Java Mission Control (JMC)]:

image::jmc-server-dump.png[]

TIP: You can get more details from a Jetty's `QueuedThreadPool` dump by enabling detailed dumps via `queuedThreadPool.setDetailedDump(true)`.
Find the `Server` MBean in the MBean Tree, under `org.eclipse.jetty.server:type=server,id=0`.
Then click on the "Operations" tab, select the `dump()` operation, and then click the `Execute` button.
In the bottom panel you will see the result of the invocation, that you can copy into a text editor and save to your file system.

[CAUTION]
====
Taking a Jetty Server Dump is a relatively expensive operation, as it dumps the state of all connections (which can be thousands), and the state of all threads.

The result of the invocation may produce a large string, possibly few MiB, that may impact the server memory usage.

Furthermore, dumping the state of the I/O Jetty components takes a little CPU time off the handling of the actual I/O, possibly slowing it down temporarily.

While the slow-down caused by taking the Jetty Server Dump may be noticeable on highly loaded systems, it is typically a very small price to pay to obtain the information about the Jetty state that may be critical to the resolution of an issue.
====

[NOTE]
====
The format of the Jetty Server Dump output is subject to change at any time, as Jetty developers modify the Jetty code and decide to include more state, or remove state that is no longer relevant.

The Jetty Server Dump is organized in a tree whose structure is similar to the runtime Jetty component tree.

At the end of the dump output there is a legend that explains the type of tree node: whether it is a node that represent a _managed_ component, or an _array_ node (or a _map_ node) that represent some component state, etc.
====

[[dump-detailed]]
== Detailed ThreadPool Information

By default, the dump of the thread pool will only dump the topmost stack frame of each thread.
It is possible to configure the thread pool to dump the whole stack trace for each thread; while this may be a little more expensive, it provides complete information about the state of each thread, which may be important to diagnose the issue.

See the xref:operations-guide:modules/standard.adoc#threadpool[`threadpool` module] configuration for more information about how to configure the thread pool to dump detailed thread information.

Detailed thread pool information can also be turned on/off on-the-fly via JMX, by finding the `ThreadPool` MBean under `org.eclipse.jetty.util.thread:type=queuedthreadpool,id=0`, then selecting the `detailedDump` attribute and setting it to `true`. You can now perform the `Server.dump()` operation as explained above, and then set `detailedDump` back to `false`.

[[dump-example]]
== Dump Example

Below you can find a simple example of a Jetty Server Dump, with annotations for the principal components:

[jetty.small%nowrap]
....
[jetty]
setupArgs=--add-modules=http
args=jetty.http.selectors=1 jetty.http.acceptors=1 jetty.threadPool.minThreads=4 jetty.server.dumpAfterStart=true
delete=^[0-9]\{4}
callouts= <$N>,Server@,= QueuedThreadPool,HandlerCollection@,= ServerConnector,ManagedSelector@,keys @,startJarLoader@,unmanaged
....
ifdef::run-jetty[]
<1> The `Server` instance at the root of the tree
<2> The thread pool component
<3> The root of the `Handler` structure
<4> The connector listening on port `8080` for the HTTP/1.1 protocol
<5> A selector component that manages connections
<6> The connections currently managed by the selector component
<7> The server `ClassLoader` and its classpath
<8> The legend for the dump nodes
endif::[]