Introduce `CompositeHttpHeadersBase` to replace HttpHeadersUtil.merge() #5340

injae-kim · 2023-12-08T13:45:47Z

related issue: #5146

Motivation:

armeria/core/src/main/java/com/linecorp/armeria/internal/common/HttpHeadersUtil.java

Line 46 in 3595dac

public static ResponseHeaders mergeResponseHeaders(ResponseHeaders headers,

On Provide a way to composite two headers #5146, we found that HttpHeadersUtil.mergeHeaders(..) occupies a large part of CPU cycles due to expensive deep-copy operations

📌 PRs Overview:

✅ (this PR) Introduce CompositeHttpHeadersBase to replace HttpHeadersUtil.mergeHeaders(..)
Introduce CompositeHttpBeaderBuilder builder
Introduce DefaultCompositeRequest, ResponseHeaders
(finally) Replace HttpHeadersUtil.mergeHeaders(..) to Composite*Headers

Modifications:

Introduce CompositeHttpHeadersBase and CompositeStringMultiMap to replace HttpHeadersUtil.merge()

Result:

Now we have CompositeHttpHeadersBase so can implement DefaultCompositeRequest, ResponseHeaders

github-actions · 2023-12-08T14:01:52Z

🔍 Build Scan® (commit: `e9fcd38`)

Job name	Status	Build Scan®
build-windows-latest-jdk-19	✅	https://ge.armeria.dev/s/fagki3cgeavso
build-self-hosted-unsafe-jdk-8	✅	https://ge.armeria.dev/s/7zm72vtzosdem
build-self-hosted-unsafe-jdk-19-snapshot-blockhound	✅	https://ge.armeria.dev/s/ry56m32qmgk56
build-self-hosted-unsafe-jdk-17-min-java-17-coverage	✅	https://ge.armeria.dev/s/w5ttm4bkqh7jm
build-self-hosted-unsafe-jdk-17-min-java-11	❌ (failure)	https://ge.armeria.dev/s/qevvxewiy5ptw
build-self-hosted-unsafe-jdk-17-leak	✅	https://ge.armeria.dev/s/cf22vfdohvxbq
build-self-hosted-unsafe-jdk-11	✅	https://ge.armeria.dev/s/cmfbbdk3iuo4e
build-macos-12-jdk-19	✅	https://ge.armeria.dev/s/zkj54c54wbns4

codecov · 2023-12-08T14:40:13Z

Codecov Report

Attention: Patch coverage is 84.47552% with 111 lines in your changes are missing coverage. Please review.

Project coverage is 74.21%. Comparing base (75d5af1) to head (e9fcd38).
Report is 108 commits behind head on main.

Files	Patch %	Lines
...necorp/armeria/common/CompositeStringMultimap.java	84.67%	52 Missing and 45 partials ⚠️
...ecorp/armeria/common/CompositeHttpHeadersBase.java	82.92%	9 Missing and 5 partials ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #5340       +/-   ##
===========================================
+ Coverage        0   74.21%   +74.21%     
- Complexity      0    21119    +21119     
===========================================
  Files           0     1803     +1803     
  Lines           0    77283    +77283     
  Branches        0     9948     +9948     
===========================================
+ Hits            0    57354    +57354     
- Misses          0    15254    +15254     
- Partials        0     4675     +4675

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

injae-kim · 2023-12-11T11:08:14Z

core/src/main/java/com/linecorp/armeria/common/CompositeStringMultimap.java

+    @Nullable
+    private StringMultimap<IN_NAME, NAME> appender;
+    @Nullable
+    private List<StringMultimap<IN_NAME, NAME>> delegates;
+
+    private final Set<IN_NAME> removed = new HashSet<>();
+    private final Map<IN_NAME, Integer> removedSizeCache = new HashMap<>();


FYI) to handle modifying operations (add, set, remove) efficiently without expensive copy, I apply append-only data structure.

add() only on appender, remove() -> just mark as removed on removed (like tombstone on kafka). 🙇

jrhee17

Thanks for the PR, left a preliminary comment 🙇

jrhee17 · 2023-12-20T15:15:13Z

core/src/main/java/com/linecorp/armeria/common/CompositeStringMultimap.java

+    }
+
+    @Override
+    public Iterator<Entry<NAME, String>> iterator() {


Question) To match the current merge logic, should this loop deduplicate such that only one key is returned?

i.e.

final CompositeStringMultimap<CharSequence, AsciiString> headers = new CompositeHttpHeadersBase(HttpHeaders.of("k1", "v1"), HttpHeaders.of("k1", "v2")); for (Entry<AsciiString, String> entry: headers) { System.out.println(entry); // ("k1", "v2") only should be printed }

ditto for the other iterating methods (forEach, forEachValue, etc..)

armeria/core/src/main/java/com/linecorp/armeria/internal/common/HttpHeadersUtil.java

Lines 46 to 48 in d328e27

public static ResponseHeaders mergeResponseHeaders(ResponseHeaders headers,

HttpHeaders additionalHeaders,

HttpHeaders defaultHeaders,

armeria/core/src/main/java/com/linecorp/armeria/internal/common/HttpHeadersUtil.java

Lines 64 to 66 in d328e27

for (AsciiString name : defaultHeaders.names()) {

if (!isPseudoHeader(name) && !builder.contains(name)) {

defaultHeaders.forEachValue(name, value -> builder.add(name, value));

Oh you're right! based on current merge logic, we don't add duplicated key on mergeHeaders(). I miss it 😅

I think CompositeHeaders can support merge (remove duplicated key) and just composite options.
(or maybe we can explicitly introduce CompositeHeader and MergedHeader class separately,
or Iwe can just simply introduce additionalHeaders, defaultHeaders field on CompositeHeader!)

Anyway, I'll think more about this design and share to you soon~! 🙇

// 1) Merge headers same with HttpHeadersUtil.merge(additionalHeaders, headers, defaultHeaders) CompositeHeader additionals = new CompositeHeader(add1, add2, ..); for (Name name : additional.names() { if (isPsedoHeader(name) || ADDITIONAL_REQUEST_HEADER_DISALLOWED_LIST.contains(name)) { additional.remove(name) // Only mark as removed, without deep copy } } .. // Can composite default and internal header like this without deep copy HttpHeaders internals = HttpHeaders.of(HttpHeaderNames.SERVER, ArmeriaHttpUtil.SERVER_HEADER); CompositeHeader defaults = new CompositeHeader(default1, internals); // 2) Also, can composite headers CompositeHeader headers = new CompositeHeader(header1, header2, ..);

✅ To respect current HttpHeadersUtil.merge(additionals, headers, defaults) logic,
I make CompositeHeader can merge additional, header, default too with same logic, without deep copy~!

minwoox

Sorry about the late review. 😓
Basically, looks good and left some suggestions. 👍

minwoox · 2024-01-10T10:58:30Z

core/src/main/java/com/linecorp/armeria/common/CompositeStringMultimap.java

+    CompositeStringMultimap(@Nullable CompositeStringMultimap<IN_NAME, NAME> additionals,
+                            List<StringMultimap<IN_NAME, NAME>> parents,
+                            @Nullable CompositeStringMultimap<IN_NAME, NAME> defaults,


It seems like the additionals and defaults CompositeStringMultimaps are tightly coupled to the logic that merges response headers and I think they complicates this class.
How about removing those and just taking the list of StringMultimap? The additionalHeaders, headers, and defaultHeaders are specified in that order so that the headers in additionalHeaders take precedence.
new CompositeHttpHeadersBase(additionalHeaders, headers, defaultHeaders)

armeria/core/src/main/java/com/linecorp/armeria/internal/common/HttpHeadersUtil.java

Lines 46 to 48 in dc2bdc6

public static ResponseHeaders mergeResponseHeaders(ResponseHeaders headers,

HttpHeaders additionalHeaders,

HttpHeaders defaultHeaders,

How about removing those and just taking the list of StringMultimap?

✅ updated, nice suggestion! this way looks much simpler 👍

minwoox · 2024-01-10T11:02:22Z

core/src/main/java/com/linecorp/armeria/common/CompositeHttpHeadersBase.java

+            if (contentLength < 0) {
+                contentLength = 0;
+            }
+            contentLength += length;


I'm not sure about aggregating all the Content-Length headers as it's unlikely that it accurately represents the actual content length. How about using the length of the first encountered header as we do for others?

How about using the length of the first encountered header as we do for others?

✅ updated! agree with this way.

Similarly, I'm not sure how we should handle on isEndOfStream case 🤔

e.g. If only some of the parents are endOfStream: true, what should we return on isEndOfStream()?

PTAL 🙇

That's a good question.
Based on previous usage, simply using the flag of the first header may not be sufficient.

I see two potential options:

If there's a headers with isEndOfStream`` set to true``, consider the composite headers as true as well. This approach seems reasonable, as the composite headers should reflect the end-of-stream status if any of them does.

Ignore the isEndOfStream flag in all headers objects and allow the user to specify it after composition.

What do you think of it?
Of course, others might have different opinions. cc @line/dx

If there's a headers with isEndOfStream set to true, consider the composite headers as true as well.

I think option1 looks better so updated code and test~! thank you! 🙇

core/src/main/java/com/linecorp/armeria/common/CompositeHttpHeadersBase.java

minwoox · 2024-01-10T11:03:06Z

core/src/main/java/com/linecorp/armeria/common/CompositeHttpHeadersBase.java

+            return true;
+        }
+
+        if (!(o instanceof HttpHeaderGetters)) {


Is this intentional not using CompositeHttpHeadersBase?

// CompositeStingMultimap if (that instanceof CompositeStringMultimap || that instanceof StringMultimap) { // allow StringMultimap to compare too for (NAME name : names()) { if (!getAll((IN_NAME) name.toString()) .equals(that.getAll((IN_NAME) name.toString()))) { return false; } }

Aha right. on CompositeStingMultimap#equal, I think we may can compare CompositeStingMultimap with StringMultimap too.

So i intended to just check !(o instanceof HttpHeaderGetters) here to allow both CompositeStingMultimap and StringMultimap.

Hmm should we allow only o instanceof CompositeHttpHeadersBase to check equality here?
I'm not sure so please share your opinion~! thanks 🙇

I think it's okay because they are equal if they have the same headers no matter what the type is.
We have the similar case for ByteArrayHttpData:

armeria/core/src/main/java/com/linecorp/armeria/common/ByteArrayHttpData.java

Lines 78 to 83 in 2aa149b

final HttpData that = (HttpData) o;

if (length() != that.length()) {

return false;

}

return Arrays.equals(array(), that.array());

I got it! I added related test case :) (compare equality with CompositeHeader and just Header)

injae-kim · 2024-02-01T06:22:12Z

Really sorry for late, I'll update this PR within this weekend 🙇

minwoox · 2024-02-22T01:00:10Z

core/src/main/java/com/linecorp/armeria/common/CompositeStringMultimap.java

+ *
+ * @see CompositeHttpHeadersBase
+ */
+abstract class CompositeStringMultimap<IN_NAME extends CharSequence, NAME extends IN_NAME>


I think is should be Multi'M'ap.

Oh I just copy&paste it from StringMultimap. should we use Multi'M'ap here?

Oh, I don't know what I was thinking when I made that comment. Sorry. Let's just leave it as it is.

minwoox · 2024-03-12T07:12:59Z

core/src/main/java/com/linecorp/armeria/common/CompositeHttpHeadersBase.java

+        super(from(parents), DEFAULT_APPENDER_SUPPLIER);
+    }
+
+    CompositeHttpHeadersBase(@Nullable List<HttpHeaderGetters> additionals,


Considering the current usage, which involves three different headers in HttpHeadersUtil, it seems that this constructor isn't necessary. Do you have a specific reason for adding it?

in HttpHeadersUtil, it seems that this constructor isn't necessary.

armeria/core/src/main/java/com/linecorp/armeria/internal/common/HttpHeadersUtil.java

Lines 57 to 62 in c6b8906

for (AsciiString name : additionalHeaders.names()) {

if (!isPseudoHeader(name)) {

builder.remove(name);

additionalHeaders.forEachValue(name, value -> builder.add(name, value));

}

}

If we don't have this constructor, how can we pass additioanls headers to CompositeStringMultimap? 🤔

armeria/core/src/main/java/com/linecorp/armeria/common/AbstractHttpHeadersBuilder.java

Lines 32 to 40 in c6b8906

@Override

final HttpHeadersBase newSetters(int sizeHint) {

return new HttpHeadersBase(sizeHint);

}

@Override

final HttpHeadersBase newSetters(HttpHeadersBase parent, boolean shallowCopy) {

return new HttpHeadersBase(parent, shallowCopy);

}

armeria/core/src/main/java/com/linecorp/armeria/common/DefaultHttpHeadersBuilder.java

Lines 24 to 30 in c6b8906

DefaultHttpHeadersBuilder(HttpHeadersBase parent) {

super(parent);

}

@Override

public HttpHeaders build() {

final HttpHeadersBase delegate = delegate();

Also, this constructor is needed for next PR, to add CompositeHttpHeaderBuilder similar with HttpHeaderBuilder!
It uses new HttpHeaderBase(..) constuructor.

If we don't have this constructor, how can we pass additioanls headers to CompositeStringMultimap?

CompositeHttpHeadersBase(HttpHeaderGetters... parents) {...}

This constructor accepts multiple HttpHeaders. Can't users simply specify the additional headers that take precedence as the first argument?

/** * Merges the given {@link ResponseHeaders}. The headers have priority in the following order. * <pre>{@code * additional headers (highest priority) > headers > default headers > framework headers (lowest priority) * }</pre> */ public static ResponseHeaders mergeResponseHeaders(ResponseHeaders headers, HttpHeaders additionalHeaders, HttpHeaders defaultHeaders) { ... new CompositeHttpHeadersBase(additionalHeaders, headers, defaultHeaders); ... }

this constructor is needed for next PR, to add CompositeHttpHeaderBuilder similar with HttpHeaderBuilder!
It uses new HttpHeaderBase(..) constructor

I might miss this. Could you share an example of how it will be used?

minwoox · 2024-03-12T07:25:58Z

core/src/main/java/com/linecorp/armeria/common/CompositeHttpHeadersBase.java

+
+    @Override
+    public boolean isEndOfStream() {
+        for (StringMultimapGetters<CharSequence, AsciiString> delegate : delegates()) {


Shouldn't we check the appender first?
Additionally, we need to ensure that the appender isn't created if it's null.

#5340 (comment) If there's a headers with isEndOfStream set to true, consider the composite headers as true as well.

As we discussed like above, I think it's not mandatory to always check appender first on isEndOfStream !

Cause if one of headers is end-of-stream, we return composite headers' isEndOfStream as true.
For optimization, we can check appender first 😄

Additionally, we need to ensure that the appender isn't created if it's null.

protected final StringMultimap<IN_NAME, NAME> appender() { if (appender != null) { return appender; } appender = requireNonNull(appenderSupplier.get(), "appender"); final ImmutableList.Builder<StringMultimapGetters<IN_NAME, NAME>> builder = ImmutableList.builder(); // 👈👈 If appender is created, add appender on delegates. delegates = builder.addAll(delegates()) .add(appender) .build(); return appender; } protected final List<StringMultimapGetters<IN_NAME, NAME>> delegates() { if (delegates != null) { // 👈👈 if appender is created or delegates() is called before, return it. return delegates; } }

Nice point! I considered this case and on delegates(), it doesn't contain appender if appender is not created yet~!

So we don't have to check appender is null or not when using delegates()!

As we discussed like above, I think it's not mandatory to always check appender first on isEndOfStream !

I apologize for my previous comment. It doesn't contain all the information
I thought we still needed to check the appender if endOfStream is set after the CompositeHttpHeadersBase is created.
I imagined a case something like:

CompositeHttpHeadersBase headers = new CompositeHttpHeadersBase(additionalHeaders, endOfStreamTrueHeaders); headers.endOfStream(false);

Additionally, we need to ensure that the appender isn't created if it's null.

What I meant was that we need to ensure that the appender is not created if we change the logic to check for the appender first. 😉

injae-kim requested review from ikhoon, jrhee17, minwoox and trustin as code owners December 8, 2023 13:45

injae-kim force-pushed the composite-http-headers branch from d3ab464 to 8a080c9 Compare December 8, 2023 13:57

injae-kim force-pushed the composite-http-headers branch from 8a080c9 to 1c3776f Compare December 8, 2023 14:06

injae-kim commented Dec 11, 2023

View reviewed changes

jrhee17 reviewed Dec 20, 2023

View reviewed changes

injae-kim force-pushed the composite-http-headers branch from a5d6074 to b088bae Compare December 27, 2023 06:27

minwoox added the improvement label Jan 10, 2024

minwoox added this to the 1.27.0 milestone Jan 10, 2024

minwoox reviewed Jan 10, 2024

View reviewed changes

ikhoon modified the milestones: 1.27.0, 1.28.0 Jan 16, 2024

injae-kim force-pushed the composite-http-headers branch 2 times, most recently from 3f7688f to 1480260 Compare February 4, 2024 17:05

minwoox reviewed Feb 22, 2024

View reviewed changes

injae-kim force-pushed the composite-http-headers branch from 1480260 to d291c12 Compare March 9, 2024 07:36

injae-kim added 5 commits March 9, 2024 16:51

Introduce CompositeHttpHeadersBase to replace HttpHeadersUtil.merge()

259c16f

Make compositeHeader can merge additional and default headers

d7275e8

Take List<StringMultimap> additionals, defaults

aaf426d

Remove complicate aggregation and just use first encountered header

18f4920

Return isEndOfStream as true if some header's endOfStream set true

a2c2728

injae-kim force-pushed the composite-http-headers branch from bb5ef93 to f35a37f Compare March 9, 2024 07:52

Add equals with HttpHeaderGetters and endOfStream test cases

e9fcd38

injae-kim force-pushed the composite-http-headers branch from f35a37f to e9fcd38 Compare March 9, 2024 08:17

minwoox reviewed Mar 12, 2024

View reviewed changes

jrhee17 modified the milestones: 1.28.0, 1.29.0 Apr 2, 2024

github-actions bot added the Stale label May 3, 2024

minwoox modified the milestones: 1.29.0, 1.30.0 May 21, 2024

github-actions bot removed the Stale label May 22, 2024

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Introduce `CompositeHttpHeadersBase` to replace HttpHeadersUtil.merge() #5340

Introduce `CompositeHttpHeadersBase` to replace HttpHeadersUtil.merge() #5340

injae-kim commented Dec 8, 2023

github-actions bot commented Dec 8, 2023 •

edited

codecov bot commented Dec 8, 2023 •

edited

injae-kim Dec 11, 2023

jrhee17 left a comment

jrhee17 Dec 20, 2023 •

edited

injae-kim Dec 22, 2023 •

edited

injae-kim Dec 27, 2023

minwoox left a comment

minwoox Jan 10, 2024

injae-kim Feb 4, 2024

minwoox Jan 10, 2024

injae-kim Feb 4, 2024 •

edited

minwoox Feb 22, 2024

injae-kim Mar 9, 2024 •

edited

minwoox Jan 10, 2024

injae-kim Feb 4, 2024 •

edited

minwoox Feb 21, 2024

injae-kim Mar 9, 2024

injae-kim commented Feb 1, 2024

minwoox Feb 22, 2024

injae-kim Mar 9, 2024

minwoox Mar 12, 2024

minwoox Mar 12, 2024

injae-kim Mar 23, 2024

minwoox Mar 25, 2024

minwoox Mar 12, 2024

injae-kim Mar 23, 2024 •

edited

injae-kim Mar 23, 2024

minwoox Mar 25, 2024 •

edited

	public static ResponseHeaders mergeResponseHeaders(ResponseHeaders headers,
	HttpHeaders additionalHeaders,
	HttpHeaders defaultHeaders,

	for (AsciiString name : defaultHeaders.names()) {
	if (!isPseudoHeader(name) && !builder.contains(name)) {
	defaultHeaders.forEachValue(name, value -> builder.add(name, value));

	final HttpData that = (HttpData) o;
	if (length() != that.length()) {
	return false;
	}

	return Arrays.equals(array(), that.array());

	for (AsciiString name : additionalHeaders.names()) {
	if (!isPseudoHeader(name)) {
	builder.remove(name);
	additionalHeaders.forEachValue(name, value -> builder.add(name, value));
	}
	}

	@Override
	final HttpHeadersBase newSetters(int sizeHint) {
	return new HttpHeadersBase(sizeHint);
	}

	@Override
	final HttpHeadersBase newSetters(HttpHeadersBase parent, boolean shallowCopy) {
	return new HttpHeadersBase(parent, shallowCopy);
	}

	DefaultHttpHeadersBuilder(HttpHeadersBase parent) {
	super(parent);
	}

	@Override
	public HttpHeaders build() {
	final HttpHeadersBase delegate = delegate();

Introduce CompositeHttpHeadersBase to replace HttpHeadersUtil.merge() #5340

Are you sure you want to change the base?

Introduce CompositeHttpHeadersBase to replace HttpHeadersUtil.merge() #5340

Conversation

injae-kim commented Dec 8, 2023

Motivation:

📌 PRs Overview:

Modifications:

Result:

github-actions bot commented Dec 8, 2023 • edited

🔍 Build Scan® (commit: e9fcd38)

codecov bot commented Dec 8, 2023 • edited

Codecov Report

Choose a reason for hiding this comment

jrhee17 left a comment

Choose a reason for hiding this comment

jrhee17 Dec 20, 2023 • edited

Choose a reason for hiding this comment

injae-kim Dec 22, 2023 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

minwoox left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

injae-kim Feb 4, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

injae-kim Mar 9, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

injae-kim Feb 4, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

injae-kim commented Feb 1, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

injae-kim Mar 23, 2024 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

minwoox Mar 25, 2024 • edited

Choose a reason for hiding this comment

Introduce `CompositeHttpHeadersBase` to replace HttpHeadersUtil.merge() #5340

Introduce `CompositeHttpHeadersBase` to replace HttpHeadersUtil.merge() #5340

github-actions bot commented Dec 8, 2023 •

edited

🔍 Build Scan® (commit: `e9fcd38`)

codecov bot commented Dec 8, 2023 •

edited

jrhee17 Dec 20, 2023 •

edited

injae-kim Dec 22, 2023 •

edited

injae-kim Feb 4, 2024 •

edited

injae-kim Mar 9, 2024 •

edited

injae-kim Feb 4, 2024 •

edited

injae-kim Mar 23, 2024 •

edited

minwoox Mar 25, 2024 •

edited