8319055: JCMD should not buffer the whole output of commands #23405

alexmenkov · 2025-01-31T23:23:36Z

The fix implements streaming output support for attach protocol.
See JBS issue for evaluation, summary of the changes in the 1st comment.
Testing: tier1..4,hs-tier5-svc

Progress

Change must be properly reviewed (1 review required, with at least 1 Reviewer)
Change must not contain extraneous whitespace
Commit message must refer to an issue

Issue

JDK-8319055: JCMD should not buffer the whole output of commands (Enhancement - P4)

Reviewers

Johan Sjölen (@jdksjolen - Reviewer) Review applies to 2a3d4346
Thomas Stuefe (@tstuefe - Reviewer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/jdk.git pull/23405/head:pull/23405
$ git checkout pull/23405

Update a local copy of the PR:
$ git checkout pull/23405
$ git pull https://git.openjdk.org/jdk.git pull/23405/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 23405

View PR using the GUI difftool:
$ git pr show -t 23405

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/jdk/pull/23405.diff

Using Webrev

Link to Webrev Comment

alexmenkov · 2025-01-31T23:24:02Z

The changes:
Attach API protocol:

attach operation options are added for ATTACH_API_V2 requests.
The options are passed as part of command name; old clients old clients enquere command without options, new client adds options only if target VM support them, so there is no need to bump attach api version;
"getversion" command supports new "options" argument. with this argument VM reports supported options.
The only supported option for now is "streaming" which allows client to control streaming/buffered output mode.

AttachListener:

implemented attachStream class to support both streaming and buffered output;
all attach command handler are updated to use attachStream instead of bufferedStream and start streaming (if enabled) after parsing command arguments;
default streaming mode is controlled by "jdk.attach.vm.streaming" system properties; this value is used when clients does not specify "streaming" option in the request (for example when tools from older releases attach to the current VM). For safety now default value is "false" (i.e. buffered output). The plan is to make the default "true" later;
platform implementations (windows/posix) updated accordingly shared code changes;

DCmd framework:

added a way to start streaming output after command/argument parsing, but before executing the command;

Client side:

added option support;
added system property "jdk.attach.allowStreamingOutput" to set streaming mode;
For now default value is "false" (i.e. to turn on streaming mode user need something like "JAVA_TOOL_OPTIONS=-Djdk.attach.allowStreamingOutput=true jcmd")
The plan is to make the default "true" after fixing tests which fail with streaming output (or maybe add command line option for jcmd and other tools).

Test CodeHeapAnalyticsParams.java

'Compiler.CodeHeap_Analytics' jcmd command is the only command which thrown AttachOperationFailedException exception if argument is out of range (and this causes jcmd exits with exit code 1).
All other jcmd commands just writes exception message in the output.
Now 'Compiler.CodeHeap_Analytics' works like other commands, so the test is updated to expect exception message.

bridgekeeper · 2025-01-31T23:24:18Z

👋 Welcome back amenkov! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

openjdk · 2025-01-31T23:24:44Z

@alexmenkov This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8319055: JCMD should not buffer the whole output of commands

Reviewed-by: stuefe, jsjolen

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been 515 new commits pushed to the master branch:

248c373: 8351266: JFR: -XX:StartFlightRecording:report-on-exit
03ef79c: 8346470: Improve WriteBarrier JMH to have old-to-young refs
b50fe9b: 8280818: Expand bug8033699.java to iterate over all LaFs
771e160: 8351323: Parameterize compiler and linker flags for iconv
0ff1c08: 8349984: (jdeps) jdeps can use String.repeat instead of String.replaceAll
37ec796: 8351500: G1: NUMA migrations cause crashes in region allocation
4e51a8c: 8307513: C2: intrinsify Math.max(long,long) and Math.min(long,long)
7e3bc81: 8351216: ZGC: Store NUMA node count
82eb780: 8351349: GSSUtil.createSubject has outdated access control context and policy related text
c3db667: 8351542: LIBMANAGEMENT_OPTIMIZATION remove special optimization settings
... and 505 more: https://git.openjdk.org/jdk/compare/d985b31cbb5646c526e1a68a7547f26f56d37607...master

As there are no conflicts, your changes will automatically be rebased on top of these commits when integrating. If you prefer to avoid this automatic rebasing, please check the documentation for the /integrate command for further details.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

openjdk · 2025-01-31T23:25:30Z

@alexmenkov The following labels will be automatically applied to this pull request:

hotspot-runtime
serviceability

When this pull request is ready to be reviewed, an "RFR" email will be sent to the corresponding mailing lists. If you would like to change these labels, use the /label pull request command.

mlbridge · 2025-01-31T23:39:27Z

Webrevs

AlanBateman · 2025-02-01T08:14:15Z

Do you expect there will be follow up work at some point for the commands that upcall to Java and produce a byte[] to return to the tool?

alexmenkov · 2025-02-03T20:09:14Z

Do you expect there will be follow up work at some point for the commands that upcall to Java and produce a byte[] to return to the tool?

I don't see much need in the functionality now (streaming output is useful for commands which produce lengthy output or take long time to execute), but we can add it if desired - need to wrap native writer to OutputStream and use the stream by java handlers (instead of byte buffer) to write command output.

AlanBateman · 2025-02-04T06:06:18Z

Do you expect there will be follow up work at some point for the commands that upcall to Java and produce a byte[] to return to the tool?

I don't see much need in the functionality now (streaming output is useful for commands which produce lengthy output or take long time to execute), but we can add it if desired - need to wrap native writer to OutputStream and use the stream by java handlers (instead of byte buffer) to write command output.

Thread.dump_to_file is an example that needs this. There will be many more in the future. So while not for this PR we should we thinking about a follow-up to put in the infrastructure to support this.

tstuefe · 2025-02-04T07:01:52Z

@alexmenkov Oh, this is great, many thanks for doing this work! Good explanations too. I had a cursory view over the changes. Will do a more thorough review, and play around with this, when I am back home.

With streaming, what is the expected behavior if a command takes too long? Jcmd stops with timeout, and the hotspot JVM then discards the results?

Unless the tests are easy to fix, I don't see any problem with just running them in old buffered mode.

jdksjolen

Hi Alex,

Thank you for this. It's very good to have this feature. I'm still reading the code, but I'm submitting these comments as a first round.

I didn't really understand what makes this streaming. The old protocol first sent out the result, and then the data, has this changed?

Thanks,
Johan

src/hotspot/share/services/diagnosticFramework.cpp

src/hotspot/share/services/attachListener.cpp

alexmenkov · 2025-02-04T19:48:29Z

Do you expect there will be follow up work at some point for the commands that upcall to Java and produce a byte[] to return to the tool?

I don't see much need in the functionality now (streaming output is useful for commands which produce lengthy output or take long time to execute), but we can add it if desired - need to wrap native writer to OutputStream and use the stream by java handlers (instead of byte buffer) to write command output.

Thread.dump_to_file is an example that needs this. There will be many more in the future. So while not for this PR we should we thinking about a follow-up to put in the infrastructure to support this.

We have JDK-8336723 to implement streaming output for commands implemented in java. This PR is lower level and provides functionality required for JDK-8336723

alexmenkov · 2025-02-04T20:04:20Z

With streaming, what is the expected behavior if a command takes too long? Jcmd stops with timeout, and the hotspot JVM then discards the results?

When a tool is run interactively (I mean not as part of a testing) there is no timeouts - the tools forward any output from the target VM to stdout until communication channel (socket/pipe) is closed.
With streaming the tool just prints command output earlier.
If jcmd is killed while the command is in progress, the VM completes the command anyway (we don't have a way to stop it), results are discarded.

Unless the tests are easy to fix, I don't see any problem with just running them in old buffered mode.

Yes, this is the plan for the next step - fix the tests of disable streaming output.

alexmenkov · 2025-02-04T20:12:44Z

I didn't really understand what makes this streaming. The old protocol first sent out the result, and then the data, has this changed?

The protocol works as before.
All command handlers are updated to set the result code earlier (after argument parsing, but before command execution).
After the result code is set, it's sent to the tool and the rest of output (data) are send directly to the tool without buffering.

test/hotspot/jtreg/serviceability/attach/AttachAPIv2/StreamingOutputTest.java

jdksjolen · 2025-02-10T18:25:58Z

I didn't really understand what makes this streaming. The old protocol first sent out the result, and then the data, has this changed?

The protocol works as before. All command handlers are updated to set the result code earlier (after argument parsing, but before command execution). After the result code is set, it's sent to the tool and the rest of output (data) are send directly to the tool without buffering.

Aha, I understand. Before, result code meant "Parsing OK, Command Execution OK", now it means "Parsing OK, starting Command Execution". That's fine by me.

Edit: I'll perform another review round soon!

jdksjolen

Hi,

Thanks for waiting.

I understand, I think, more or less what this does on the VM side now. I think that we can simplify the code for the cases when we set an error and don't write any error message.

src/hotspot/os/posix/attachListener_posix.cpp

src/hotspot/os/windows/attachListener_windows.cpp

src/hotspot/share/services/attachListener.cpp

jdksjolen

Hi Alex,

I'm generally happy with this code. I've got a few more comments, but nothing major.

src/hotspot/share/services/attachListener.cpp

src/jdk.attach/share/classes/sun/tools/attach/HotSpotVirtualMachine.java

jdksjolen

Hi,

This looks good to me now. Thank you!

tstuefe

+1

alexmenkov · 2025-03-11T03:13:17Z

Added ThreadBlockInVM in complete() method to do the same as old implementation.
It solves macos-aarch64 failure of test/hotspot/jtreg/runtime/NMT/VirtualAllocCommitUncommitRecommit.java test in GHA

alexmenkov · 2025-03-12T19:30:06Z

@jdksjolen , @tstuefe can you re-review with the last commit please

tstuefe

Looks still ok

alexmenkov · 2025-03-13T20:04:59Z

Thanks for the review
/integrate

openjdk · 2025-03-13T20:05:50Z

Going to push as commit cd1be91.
Since your change was applied there have been 515 commits pushed to the master branch:

248c373: 8351266: JFR: -XX:StartFlightRecording:report-on-exit
03ef79c: 8346470: Improve WriteBarrier JMH to have old-to-young refs
b50fe9b: 8280818: Expand bug8033699.java to iterate over all LaFs
771e160: 8351323: Parameterize compiler and linker flags for iconv
0ff1c08: 8349984: (jdeps) jdeps can use String.repeat instead of String.replaceAll
37ec796: 8351500: G1: NUMA migrations cause crashes in region allocation
4e51a8c: 8307513: C2: intrinsify Math.max(long,long) and Math.min(long,long)
7e3bc81: 8351216: ZGC: Store NUMA node count
82eb780: 8351349: GSSUtil.createSubject has outdated access control context and policy related text
c3db667: 8351542: LIBMANAGEMENT_OPTIMIZATION remove special optimization settings
... and 505 more: https://git.openjdk.org/jdk/compare/d985b31cbb5646c526e1a68a7547f26f56d37607...master

Your commit was automatically rebased without conflicts.

openjdk · 2025-03-13T20:05:58Z

@alexmenkov Pushed as commit cd1be91.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.

fix

fbc1c37

openjdk bot added serviceability [email protected] hotspot-runtime [email protected] labels Jan 31, 2025

alexmenkov added 2 commits January 31, 2025 15:31

jcheck

d9b883a

jcheck

052319f

openjdk bot added the rfr Pull request is ready for review label Jan 31, 2025

jdksjolen reviewed Feb 4, 2025

View reviewed changes

feedback

96d01f7

sspitsyn reviewed Feb 6, 2025

View reviewed changes

test/hotspot/jtreg/serviceability/attach/AttachAPIv2/StreamingOutputTest.java Show resolved Hide resolved

test/hotspot/jtreg/serviceability/attach/AttachAPIv2/StreamingOutputTest.java Show resolved Hide resolved

feedback - test

d074f2f

jdksjolen reviewed Mar 5, 2025

View reviewed changes

feedback

5c87f9e

jdksjolen reviewed Mar 6, 2025

View reviewed changes

feedback

2a3d434

jdksjolen approved these changes Mar 7, 2025

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Mar 7, 2025

tstuefe approved these changes Mar 10, 2025

View reviewed changes

restore ThreadBlockInVM in complete

ed569cc

openjdk bot removed the ready Pull request is ready to be integrated label Mar 11, 2025

tstuefe approved these changes Mar 13, 2025

View reviewed changes

openjdk bot added the ready Pull request is ready to be integrated label Mar 13, 2025

openjdk bot added the integrated Pull request has been integrated label Mar 13, 2025

openjdk bot closed this Mar 13, 2025

openjdk bot removed ready Pull request is ready to be integrated rfr Pull request is ready for review labels Mar 13, 2025

alexmenkov deleted the attach_streaming branch March 13, 2025 20:11

varada1110 mentioned this pull request Mar 27, 2025

8352392: AIX: implement attach API v2 and streaming output #24177

Closed

3 tasks

8319055: JCMD should not buffer the whole output of commands #23405

8319055: JCMD should not buffer the whole output of commands #23405

Uh oh!

Conversation

alexmenkov commented Jan 31, 2025 • edited by openjdk bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Progress

Issue

Reviewers

Reviewing

Uh oh!

alexmenkov commented Jan 31, 2025

Uh oh!

bridgekeeper bot commented Jan 31, 2025

Uh oh!

openjdk bot commented Jan 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

openjdk bot commented Jan 31, 2025

Uh oh!

mlbridge bot commented Jan 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Webrevs

Uh oh!

AlanBateman commented Feb 1, 2025

Uh oh!

alexmenkov commented Feb 3, 2025

Uh oh!

AlanBateman commented Feb 4, 2025

Uh oh!

tstuefe commented Feb 4, 2025

Uh oh!

jdksjolen left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

alexmenkov commented Feb 4, 2025

Uh oh!

alexmenkov commented Feb 4, 2025

Uh oh!

alexmenkov commented Feb 4, 2025

Uh oh!

Uh oh!

Uh oh!

jdksjolen commented Feb 10, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jdksjolen left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jdksjolen left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jdksjolen left a comment

Choose a reason for hiding this comment

Uh oh!

tstuefe left a comment

Choose a reason for hiding this comment

Uh oh!

alexmenkov commented Mar 11, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

alexmenkov commented Mar 12, 2025

Uh oh!

tstuefe left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

alexmenkov commented Jan 31, 2025 •

edited by openjdk bot

Loading

openjdk bot commented Jan 31, 2025 •

edited

Loading

mlbridge bot commented Jan 31, 2025 •

edited

Loading

jdksjolen commented Feb 10, 2025 •

edited

Loading

alexmenkov commented Mar 11, 2025 •

edited

Loading

tstuefe left a comment •

edited

Loading