kazusa
/
skywalking
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Improve match performance of URI matching in 40% (#10923) 

## Conclusion
1. In matching the first rule scenario, the new implementation(mostly due to split performance), it is 40% fast than the previous implementation.
2. Quick match has a huge advantage in GC(time and cost) and memory cost(alloc, G1 Eden and G1 Old) in every case. In my 64G M1 Max env, the regex matching even causes full GC just in the benchmark, which is only running in 2:11 mins. The regex would be a disaster for product env memory.
3. The quick match only has lower performance when the number of pattern rules is less than 4, which is nearly impossible. Even for that, Regex matching costs most GC and memory alloc in a one-time match, and 3-time cost in 4 times match.

## Other Changes
1. I made the URI cache only works in AI pipeline works, as it should. This was a bug in the previous commit.
2. `/` is not required in this case, as generally, the pattern recognition should work well for the demo name as well.
This commit is contained in:
吴晟 Wu Sheng 2023-06-13 20:15:40 +08:00 committed by GitHub
parent feb3ca2b6d
commit 848583cde5
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 107 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

102
oap-server/microbench/src/main/java/org/apache/skywalking/oap/server/microbench/core/config/group/uri/RegexVSQuickMatchBenchmark.java
View File

@ -103,6 +103,16 @@ public class RegexVSQuickMatchBenchmark extends AbstractMicrobenchmark {
bh.consume(formatClass.match("service1", "/products/123"));
}
@Benchmark
public void matchFourthRegex(Blackhole bh, RegexVSQuickMatchBenchmark.RegexMatch formatClass) {
bh.consume(formatClass.match("service1", "/sales/123/2"));
}
@Benchmark
public void matchFourthQuickUriGrouping(Blackhole bh, RegexVSQuickMatchBenchmark.QuickMatch formatClass) {
bh.consume(formatClass.match("service1", "/sales/123/2"));
}
@Benchmark
public void notMatchRegex(Blackhole bh, RegexVSQuickMatchBenchmark.RegexMatch formatClass) {
bh.consume(formatClass.match("service1", "/employees/123"));
@ -126,41 +136,59 @@ public class RegexVSQuickMatchBenchmark extends AbstractMicrobenchmark {
* # Benchmark mode: Throughput, ops/time
* # Benchmark: org.apache.skywalking.oap.server.microbench.core.config.group.uri.RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping
*
Benchmark Mode Cnt Score Error Units
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping thrpt 21824009.131 ops/s
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.alloc.rate thrpt 5388.565 MB/sec
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.alloc.rate.norm thrpt 272.000 B/op
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Eden_Space thrpt 5414.783 MB/sec
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Eden_Space.norm thrpt 273.323 B/op
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Old_Gen thrpt 0.019 MB/sec
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Old_Gen.norm thrpt 0.001 B/op
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.count thrpt 186.000 counts
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.time thrpt 112.000 ms
RegexVSQuickMatchBenchmark.matchFirstRegex thrpt 52815735.190 ops/s
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.alloc.rate thrpt 7286.910 MB/sec
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.alloc.rate.norm thrpt 152.000 B/op
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Eden_Space thrpt 7306.492 MB/sec
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Eden_Space.norm thrpt 152.408 B/op
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Old_Gen thrpt 0.026 MB/sec
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Old_Gen.norm thrpt 0.001 B/op
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.count thrpt 251.000 counts
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.time thrpt 143.000 ms
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping thrpt 23265119.660 ops/s
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.alloc.rate thrpt 5914.434 MB/sec
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.alloc.rate.norm thrpt 280.000 B/op
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Eden_Space thrpt 5939.872 MB/sec
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Eden_Space.norm thrpt 281.204 B/op
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Old_Gen thrpt 0.025 MB/sec
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Old_Gen.norm thrpt 0.001 B/op
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.count thrpt 204.000 counts
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.time thrpt 108.000 ms
RegexVSQuickMatchBenchmark.notMatchRegex thrpt 10462244.647 ops/s
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.alloc.rate thrpt 26817.565 MB/sec
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.alloc.rate.norm thrpt 2824.000 B/op
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Eden_Space thrpt 26925.722 MB/sec
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Eden_Space.norm thrpt 2835.389 B/op
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Old_Gen thrpt 0.128 MB/sec
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Old_Gen.norm thrpt 0.013 B/op
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.count thrpt 925.000 counts
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.time thrpt 507.000 ms
Benchmark Mode Cnt Score Error Units
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping thrpt 28464926.797 ops/s
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.alloc.rate thrpt 6194.492 MB/sec
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.alloc.rate.norm thrpt 240.000 B/op
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Eden_Space thrpt 6222.267 MB/sec
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Eden_Space.norm thrpt 241.076 B/op
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Old_Gen thrpt 0.023 MB/sec
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.churn.G1_Old_Gen.norm thrpt 0.001 B/op
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.count thrpt 214.000 counts
RegexVSQuickMatchBenchmark.matchFirstQuickUriGrouping:·gc.time thrpt 194.000 ms
RegexVSQuickMatchBenchmark.matchFirstRegex thrpt 51679120.204 ops/s
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.alloc.rate thrpt 7130.116 MB/sec
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.alloc.rate.norm thrpt 152.000 B/op
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Eden_Space thrpt 7162.842 MB/sec
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Eden_Space.norm thrpt 152.698 B/op
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Old_Gen thrpt 0.020 MB/sec
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.churn.G1_Old_Gen.norm thrpt ≈ 10⁻³ B/op
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.count thrpt 246.000 counts
RegexVSQuickMatchBenchmark.matchFirstRegex:·gc.time thrpt 224.000 ms
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping thrpt 23359343.934 ops/s
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.alloc.rate thrpt 6106.164 MB/sec
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.alloc.rate.norm thrpt 288.000 B/op
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.churn.G1_Eden_Space thrpt 6143.526 MB/sec
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.churn.G1_Eden_Space.norm thrpt 289.762 B/op
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.churn.G1_Old_Gen thrpt 0.023 MB/sec
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.churn.G1_Old_Gen.norm thrpt 0.001 B/op
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.count thrpt 211.000 counts
RegexVSQuickMatchBenchmark.matchFourthQuickUriGrouping:·gc.time thrpt 143.000 ms
RegexVSQuickMatchBenchmark.matchFourthRegex thrpt 24074353.094 ops/s
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.alloc.rate thrpt 17999.991 MB/sec
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.alloc.rate.norm thrpt 824.000 B/op
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.churn.G1_Eden_Space thrpt 18070.905 MB/sec
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.churn.G1_Eden_Space.norm thrpt 827.246 B/op
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.churn.G1_Old_Gen thrpt 0.095 MB/sec
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.churn.G1_Old_Gen.norm thrpt 0.004 B/op
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.count thrpt 621.000 counts
RegexVSQuickMatchBenchmark.matchFourthRegex:·gc.time thrpt 934.000 ms
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping thrpt 27031477.704 ops/s
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.alloc.rate thrpt 6081.482 MB/sec
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.alloc.rate.norm thrpt 248.000 B/op
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Eden_Space thrpt 6109.321 MB/sec
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Eden_Space.norm thrpt 249.135 B/op
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Old_Gen thrpt 0.022 MB/sec
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.churn.G1_Old_Gen.norm thrpt 0.001 B/op
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.count thrpt 210.000 counts
RegexVSQuickMatchBenchmark.notMatchQuickUriGrouping:·gc.time thrpt 171.000 ms
RegexVSQuickMatchBenchmark.notMatchRegex thrpt 9368757.119 ops/s
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.alloc.rate thrpt 23999.619 MB/sec
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.alloc.rate.norm thrpt 2824.000 B/op
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Eden_Space thrpt 24087.019 MB/sec
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Eden_Space.norm thrpt 2834.284 B/op
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Old_Gen thrpt 0.114 MB/sec
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.churn.G1_Old_Gen.norm thrpt 0.013 B/op
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.count thrpt 828.000 counts
RegexVSQuickMatchBenchmark.notMatchRegex:·gc.time thrpt 896.000 ms
*/

40
oap-server/server-core/src/main/java/org/apache/skywalking/oap/server/core/config/group/EndpointNameGrouping.java
View File

@ -39,10 +39,19 @@ import org.apache.skywalking.oap.server.library.util.StringFormatGroup;
@Slf4j
public class EndpointNameGrouping {
/**
* Endpoint grouping according to local endpoint-name-grouping.yml or associated dynamic configuration.
*/
@Setter
private volatile QuickUriGroupingRule endpointGroupingRule;
/**
* Endpoint grouping according to OpenAPI specification. The OpenAPI spec file is expected to provide.
*/
@Setter
private volatile EndpointGroupingRule4Openapi endpointGroupingRule4Openapi;
/**
* AI pipeline based HTTP URI pattern recognition. The rules are automatically generated by remote AI/ML service.
*/
@Setter
private volatile QuickUriGroupingRule quickUriGroupingRule;
/**
@ -64,8 +73,9 @@ public class EndpointNameGrouping {
*
* @param serviceName service name
* @param endpointName endpoint name to be formatted.
* @return Tuple2 where the first element is the formatted name and the second element is a boolean
* represented the endpoint name is formatted or not.
* @return Tuple2 The result of endpoint name formatting.
* The first element is the formatted name or original name when not formatted.
* The second element is a boolean represented the endpoint name is formatted or not.
*/
public Tuple2<String, Boolean> format(String serviceName, String endpointName) {
Tuple2<String, Boolean> formattedName = new Tuple2<>(endpointName, Boolean.FALSE);
@ -79,23 +89,17 @@ public class EndpointNameGrouping {
if (!formattedName._2() && quickUriGroupingRule != null) {
formattedName = formatByQuickUriPattern(serviceName, endpointName);
}
if (!formattedName._2()) {
// Only URI includes '/' will be cached and formatted later.
// Otherwise it could be anything, and has no need to be formatted.
if (endpointName.indexOf("/") > -1) {
ConcurrentHashMap<String, AtomicInteger> svrHttpUris = cachedHttpUris.get(serviceName);
if (svrHttpUris == null) {
cachedHttpUris.putIfAbsent(serviceName, new ConcurrentHashMap<>());
svrHttpUris = cachedHttpUris.get(serviceName);
}
// Only cache first 2000 URIs per 30 mins.
if (svrHttpUris.size() < maxHttpUrisNumberPerService) {
final AtomicInteger cachedCount = svrHttpUris.putIfAbsent(endpointName, new AtomicInteger(1));
if (null != cachedCount) {
cachedCount.incrementAndGet();
}
ConcurrentHashMap<String, AtomicInteger> svrHttpUris = cachedHttpUris.get(serviceName);
if (svrHttpUris == null) {
cachedHttpUris.putIfAbsent(serviceName, new ConcurrentHashMap<>());
svrHttpUris = cachedHttpUris.get(serviceName);
}
// Only cache first N(determined by maxHttpUrisNumberPerService) URIs per 30 mins.
if (svrHttpUris.size() < maxHttpUrisNumberPerService) {
final AtomicInteger cachedCount = svrHttpUris.putIfAbsent(formattedName._1(), new AtomicInteger(1));
if (null != cachedCount) {
cachedCount.incrementAndGet();
}
}
}

21
oap-server/server-core/src/main/java/org/apache/skywalking/oap/server/core/config/group/uri/quickmatch/PatternTree.java
View File

@ -78,8 +78,27 @@ public class PatternTree {
current.setExpression(pattern);
}
List<String> splitByCharacter(String input, char delimiter) {
List<String> parts = new ArrayList<>();
int start = 0;
for (int i = 0; i < input.length(); i++) {
if (input.charAt(i) == delimiter) {
parts.add(input.substring(start, i));
start = i + 1;
}
}
// Add the last part if necessary
if (start < input.length()) {
parts.add(input.substring(start));
}
return parts;
}
public StringFormatGroup.FormatResult match(String uri) {
final String[] slices = uri.split("/");
final List<String> slices = splitByCharacter(uri, '/');
List<PatternToken> current = roots;
PatternToken matchedToken = null;
for (final String slice : slices) {