I'm using the Google Maps Geocode API (https://github.com/googlemaps/google-maps-services-java) in a Dataflow job. My DoFn prepares the GeoApiContext at Setup. The process element function is done like so:
public void processElement(ProcessContext c) {
String address = c.element().get("Address").toString();
String id = c.element().get("Id").toString();
Gson gson = new GsonBuilder().create();
try {
GeocodingResult[] results = GeocodingApi.newRequest(this.geocodeContext).address(address).language("pt-BR").components(ComponentFilter.country("BR")).await();
if(results.length == 0) {
TableRow outputRow = new TableRow();
outputRow.set("Id", id);
c.output(outputRow);
} else {
for(GeocodingResult r : results) {
TableRow outputRow = convertTableRow(gson.toJson(r).toString());
outputRow.set("Id", id);
c.output(outputRow);
}
}
} catch(ApiException e) {
LOGGER.error("ApiException on address: {}", address, e);
} catch(InterruptedException e) {
LOGGER.error("InterruptedException on address: {}", address, e);
} catch(IOException e) {
LOGGER.error("IOException on address: {}", address, e);
}
}
This code worked fine locally, but when deployed to dataflow it throws a network error:
exception: "java.net.ConnectException: Failed to connect to maps.googleapis.com/2607:f8b0:4001:c05:0:0:0:5f:443
at okhttp3.internal.connection.RealConnection.connectSocket(RealConnection.java:265)
at okhttp3.internal.connection.RealConnection.connect(RealConnection.java:183)
at okhttp3.internal.connection.ExchangeFinder.findConnection(ExchangeFinder.java:224)
at okhttp3.internal.connection.ExchangeFinder.findHealthyConnection(ExchangeFinder.java:108)
at okhttp3.internal.connection.ExchangeFinder.find(ExchangeFinder.java:88)
at okhttp3.internal.connection.Transmitter.newExchange(Transmitter.java:169)
at okhttp3.internal.connection.ConnectInterceptor.intercept(ConnectInterceptor.java:41)
at okhttp3.internal.http.RealInterceptorChain.proceed(RealInterceptorChain.java:142)
at okhttp3.internal.http.RealInterceptorChain.proceed(RealInterceptorChain.java:117)
at okhttp3.internal.cache.CacheInterceptor.intercept(CacheInterceptor.java:94)
at okhttp3.internal.http.RealInterceptorChain.proceed(RealInterceptorChain.java:142)
at okhttp3.internal.http.RealInterceptorChain.proceed(RealInterceptorChain.java:117)
at okhttp3.internal.http.BridgeInterceptor.intercept(BridgeInterceptor.java:93)
at okhttp3.internal.http.RealInterceptorChain.proceed(RealInterceptorChain.java:142)
at okhttp3.internal.http.RetryAndFollowUpInterceptor.intercept(RetryAndFollowUpInterceptor.java:88)
at okhttp3.internal.http.RealInterceptorChain.proceed(RealInterceptorChain.java:142)
at okhttp3.internal.http.RealInterceptorChain.proceed(RealInterceptorChain.java:117)
at okhttp3.RealCall.getResponseWithInterceptorChain(RealCall.java:229)
at okhttp3.RealCall$AsyncCall.execute(RealCall.java:172)
at okhttp3.internal.NamedRunnable.run(NamedRunnable.java:32)
at java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1128)
at java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:628)
at java.base/java.lang.Thread.run(Thread.java:834)
Caused by: java.net.ConnectException: Network is unreachable (connect failed)
at java.base/java.net.PlainSocketImpl.socketConnect(Native Method)
at java.base/java.net.AbstractPlainSocketImpl.doConnect(AbstractPlainSocketImpl.java:399)
at java.base/java.net.AbstractPlainSocketImpl.connectToAddress(AbstractPlainSocketImpl.java:242)
at java.base/java.net.AbstractPlainSocketImpl.connect(AbstractPlainSocketImpl.java:224)
at java.base/java.net.SocksSocketImpl.connect(SocksSocketImpl.java:403)
at java.base/java.net.Socket.connect(Socket.java:591)
at okhttp3.internal.platform.Platform.connectSocket(Platform.java:130)
at okhttp3.internal.connection.RealConnection.connectSocket(RealConnection.java:263)
... 22 more
I've ensured that the VM spawned has internet access and I can even ping the maps.googleapis.com endpoint from inside the container:
USER#test-geocode-07020834-qmrj-harness-3k2l ~ $ docker container ls
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
b2fd123138aa 3a1cb7aedd54 "/opt/google/dataflo…" 6 minutes ago Up 5 minutes k8s_healthchecker_dataflow-test-geocode-07020834-qmrj-harness-3k2l_default_5648e9815f2ca5beea8b0eb945e12d1f_0
086e36c3dd23 4127911f4769 "/opt/google/dataflo…" 6 minutes ago Up 5 minutes k8s_vmmonitor_dataflow-test-geocode-07020834-qmrj-harness-3k2l_default_5648e9815f2ca5beea8b0eb945e12d1f_0
2890fa415af5 664bd8972b23 "/opt/google/dataflo…" 6 minutes ago Up 6 minutes k8s_shuffle_dataflow-test-geocode-07020834-qmrj-harness-3k2l_default_5648e9815f2ca5beea8b0eb945e12d1f_0
eea757bf6be7 gcr.io/cloud-dataflow/v1beta3/beam-java11-batch "/opt/google/dataflo…" 6 minutes ago Up 6 minutes k8s_java-batch_dataflow-test-geocode-07020834-qmrj-harness-3k2l_default_5648e9815f2ca5beea8b0eb945e12d1f_0
b636784118f5 k8s.gcr.io/pause:3.1 "/pause" 6 minutes ago Up 6 minutes k8s_POD_dataflow-test-geocode-07020834-qmrj-harness-3k2l_default_5648e9815f2ca5beea8b0eb945e12d1f_0
lucas#test-geocode-07020834-qmrj-harness-3k2l ~ $ docker exec -it eea /bin/sh
# ping maps.googleapis.com
PING maps.googleapis.com (172.217.214.95) 56(84) bytes of data.
64 bytes from 172.217.214.95: icmp_seq=1 ttl=115 time=1.08 ms
64 bytes from 172.217.214.95: icmp_seq=2 ttl=115 time=1.28 ms
64 bytes from 172.217.214.95: icmp_seq=3 ttl=115 time=1.15 ms
64 bytes from 172.217.214.95: icmp_seq=4 ttl=115 time=1.41 ms
^C
--- maps.googleapis.com ping statistics ---
4 packets transmitted, 4 received, 0% packet loss, time 3004ms
rtt min/avg/max/mdev = 1.089/1.235/1.414/0.131 ms
#
Regarding versions, I'm using the latest beam version (2.22.0) and the latest google maps version (0.14.0).
No idea what else to look at, and any help is appreciated.
UPDATE
The problem seems to be the fact that the request is done with an ipv6 address. However, GCE machines seems to have no support for ipv6 and the call simply fails without falling back to ipv4.
Considering that, there doesn't seem to be any way out of this problem:
Configuring the JVM to prefer ipv4 address can't be done with Dataflow (JVM flags are ignored)
There's also no way to customize the GCE machine (since a base Dataflow image is used)
The library doesn't seem to open any options to configure ipv4 or ipv6
Thanks
I had this exact same issue come up after upgrading from 2.17 to 2.24 and changing from Java 8 to Java 11. After trying to fix this on 2.24 and Java 11 I gave up and went back to 8 and it's working now.
I couldn't find it documented anywhere but it looks like the userAgent used is based on that -
When I build the self executable with Java 8, Dataflow shows the userAgent as
Apache_Beam_SDK_for_Java/2.24.0(JRE_8_environment)
and with Java 11 it shows Apache_Beam_SDK_for_Java/2.24.0(JDK_11_environment)
We have a client application (with 10+ years of development). Its JDK was upgraded from OpenJDK 11 to OpenJDK 14 recently. On single-CPU (hyper-threading disabled) Windows 10 setups (and inside VirtualBox machines with only one available CPU) the application starts quite slowly compared to Java 11. Furthermore, it uses 100% CPU most of the time. We could also reproduce the issue with setting the processor affinity to only one CPU (c:\windows\system32\cmd.exe /C start /affinity 1 ...).
Some measurement with starting the application and doing a query with minimal manual interaction in my VirtualBox machine:
OpenJDK 11.0.2: 36 seconds
OpenJDK 13.0.2: ~1.5 minutes
OpenJDK 13.0.2 with -XX:-UseBiasedLocking: 46 seconds
OpenJDK 13.0.2 with -XX:-ThreadLocalHandshakes: 40 seconds
OpenJDK 14: 5-6 minutes
OpenJDK 14 with -XX:-UseBiasedLocking: 3-3,5 minutes
OpenJDK 15 EA Build 20: ~4,5 minutes
Only the used JDK (and the mentioned options) has been changed. (-XX:-ThreadLocalHandshakes is not available in Java 14.)
We have tried logging what JDK 14 does with -Xlog:all=debug:file=app.txt:uptime,tid,level,tags:filecount=50.
Counting the log lines for every second seems quite smooth with OpenJDK 11.0.2:
$ cat jdk11-log/app* | grep "^\[" | cut -d. -f 1 | cut -d[ -f 2 | sort | uniq -c | sort -k 2 -n
30710 0
44012 1
55461 2
55974 3
27182 4
41292 5
43796 6
51889 7
54170 8
58850 9
51422 10
44378 11
41405 12
53589 13
41696 14
29526 15
2350 16
50228 17
62623 18
42684 19
45045 20
On the other hand, OpenJDK 14 seems to have interesting quiet periods:
$ cat jdk14-log/app* | grep "^\[" | cut -d. -f 1 | cut -d[ -f 2 | sort | uniq -c | sort -k 2 -n
7726 0
1715 5
10744 6
4341 11
42792 12
45979 13
38783 14
17253 21
34747 22
1025 28
2079 33
2398 39
3016 44
So, what's happening between seconds 1-4, 7-10 and 14-20?
...
[0.350s][7248][debug][class,resolve ] jdk.internal.ref.CleanerFactory$1 java.lang.Thread CleanerFactory.java:45
[0.350s][7248][debug][class,resolve ] jdk.internal.ref.CleanerImpl java.lang.Thread CleanerImpl.java:117
[0.350s][7248][info ][biasedlocking ] Aligned thread 0x000000001727e010 to 0x000000001727e800
[0.350s][7248][info ][os,thread ] Thread started (tid: 2944, attributes: stacksize: default, flags: CREATE_SUSPENDED STACK_SIZE_PARAM_IS)
[0.350s][6884][info ][os,thread ] Thread is alive (tid: 6884).
[0.350s][6884][debug][os,thread ] Thread 6884 stack dimensions: 0x00000000175b0000-0x00000000176b0000 (1024k).
[0.350s][6884][debug][os,thread ] Thread 6884 stack guard pages activated: 0x00000000175b0000-0x00000000175b4000.
[0.350s][7248][debug][thread,smr ] tid=7248: Threads::add: new ThreadsList=0x0000000017254500
[0.350s][7248][debug][thread,smr ] tid=7248: ThreadsSMRSupport::free_list: threads=0x0000000017253d50 is freed.
[0.350s][2944][info ][os,thread ] Thread is alive (tid: 2944).
[0.350s][2944][debug][os,thread ] Thread 2944 stack dimensions: 0x00000000177b0000-0x00000000178b0000 (1024k).
[0.350s][2944][debug][os,thread ] Thread 2944 stack guard pages activated: 0x00000000177b0000-0x00000000177b4000.
[0.351s][2944][debug][class,resolve ] java.lang.Thread java.lang.Runnable Thread.java:832
[0.351s][2944][debug][class,resolve ] jdk.internal.ref.CleanerImpl jdk.internal.misc.InnocuousThread CleanerImpl.java:135
[0.351s][2944][debug][class,resolve ] jdk.internal.ref.CleanerImpl jdk.internal.ref.PhantomCleanable CleanerImpl.java:138
[0.351s][2944][info ][biasedlocking,handshake] JavaThread 0x000000001727e800 handshaking JavaThread 0x000000000286d800 to revoke object 0x00000000c0087f78
[0.351s][2944][debug][vmthread ] Adding VM operation: HandshakeOneThread
[0.351s][6708][debug][vmthread ] Evaluating non-safepoint VM operation: HandshakeOneThread
[0.351s][6708][debug][vmoperation ] begin VM_Operation (0x00000000178af250): HandshakeOneThread, mode: no safepoint, requested by thread 0x000000001727e800
# no log until 5.723s
[5.723s][7248][info ][biasedlocking ] Revoked bias of currently-unlocked object
[5.723s][7248][debug][handshake,task ] Operation: RevokeOneBias for thread 0x000000000286d800, is_vm_thread: false, completed in 94800 ns
[5.723s][7248][debug][class,resolve ] java.util.zip.ZipFile$CleanableResource java.lang.ref.Cleaner ZipFile.java:715
[5.723s][7248][debug][class,resolve ] java.lang.ref.Cleaner jdk.internal.ref.CleanerImpl$PhantomCleanableRef Cleaner.java:220
[5.723s][7248][debug][class,resolve ] java.util.zip.ZipFile$CleanableResource java.util.WeakHashMap ZipFile.java:716
...
The second pause a little bit later:
...
[6.246s][7248][info ][class,load ] java.awt.Graphics source: jrt:/java.desktop
[6.246s][7248][debug][class,load ] klass: 0x0000000100081a00 super: 0x0000000100001080 loader: [loader data: 0x0000000002882bd0 of 'bootstrap'] bytes: 5625 checksum: 0025818f
[6.246s][7248][debug][class,resolve ] java.awt.Graphics java.lang.Object (super)
[6.246s][7248][info ][class,loader,constraints] updating constraint for name java/awt/Graphics, loader 'bootstrap', by setting class object
[6.246s][7248][debug][jit,compilation ] 19 4 java.lang.Object::<init> (1 bytes) made not entrant
[6.246s][7248][debug][vmthread ] Adding VM operation: HandshakeAllThreads
[6.246s][6708][debug][vmthread ] Evaluating non-safepoint VM operation: HandshakeAllThreads
[6.246s][6708][debug][vmoperation ] begin VM_Operation (0x000000000203ddf8): HandshakeAllThreads, mode: no safepoint, requested by thread 0x000000000286d800
[6.246s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026b0800, is_vm_thread: true, completed in 1400 ns
[6.246s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026bb800, is_vm_thread: true, completed in 700 ns
[6.246s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026ef800, is_vm_thread: true, completed in 100 ns
[6.246s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026f0800, is_vm_thread: true, completed in 100 ns
[6.246s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026f1800, is_vm_thread: true, completed in 100 ns
[6.246s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026f4800, is_vm_thread: true, completed in 100 ns
[6.247s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x0000000002768800, is_vm_thread: true, completed in 100 ns
[6.247s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x000000000276e000, is_vm_thread: true, completed in 100 ns
[6.247s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x0000000017268800, is_vm_thread: true, completed in 100 ns
[6.247s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x000000001727e800, is_vm_thread: true, completed in 800 ns
# no log until 11.783s
[11.783s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x000000000286d800, is_vm_thread: true, completed in 6300 ns
[11.783s][6708][info ][handshake ] Handshake "Deoptimize", Targeted threads: 11, Executed by targeted threads: 0, Total completion time: 5536442500 ns
[11.783s][6708][debug][vmoperation ] end VM_Operation (0x000000000203ddf8): HandshakeAllThreads, mode: no safepoint, requested by thread 0x000000000286d800
[11.783s][7248][debug][protectiondomain ] Checking package access
[11.783s][7248][debug][protectiondomain ] class loader: a 'jdk/internal/loader/ClassLoaders$AppClassLoader'{0x00000000c0058628} protection domain: a 'java/security/ProtectionDomain'{0x00000000c058b948} loading: 'java/awt/Graphics'
[11.783s][7248][debug][protectiondomain ] granted
[11.783s][7248][debug][class,resolve ] sun.launcher.LauncherHelper java.awt.Graphics LauncherHelper.java:816 (reflection)
[11.783s][7248][debug][class,resolve ] jdk.internal.reflect.Reflection [Ljava.lang.reflect.Method; Reflection.java:300
[11.783s][7248][debug][class,preorder ] java.lang.PublicMethods$MethodList source: C:\Users\example\AppData\Local\example\stable\jdk\lib\modules
...
Then the third one:
...
[14.578s][7248][debug][class,preorder ] java.lang.InheritableThreadLocal source: C:\Users\example\AppData\Local\example\stable\jdk\lib\modules
[14.578s][7248][info ][class,load ] java.lang.InheritableThreadLocal source: jrt:/java.base
[14.578s][7248][debug][class,load ] klass: 0x0000000100124740 super: 0x0000000100021a18 loader: [loader data: 0x0000000002882bd0 of 'bootstrap'] bytes: 1338 checksum: 8013ed55
[14.578s][7248][debug][class,resolve ] java.lang.InheritableThreadLocal java.lang.ThreadLocal (super)
[14.578s][7248][debug][jit,compilation ] 699 3 java.lang.ThreadLocal::get (38 bytes) made not entrant
[14.578s][7248][debug][vmthread ] Adding VM operation: HandshakeAllThreads
[14.578s][6708][debug][vmthread ] Evaluating non-safepoint VM operation: HandshakeAllThreads
[14.578s][6708][debug][vmoperation ] begin VM_Operation (0x000000000203d228): HandshakeAllThreads, mode: no safepoint, requested by thread 0x000000000286d800
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026b0800, is_vm_thread: true, completed in 1600 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026bb800, is_vm_thread: true, completed in 900 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026ef800, is_vm_thread: true, completed in 100 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026f0800, is_vm_thread: true, completed in 100 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026f1800, is_vm_thread: true, completed in 100 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x00000000026f4800, is_vm_thread: true, completed in 0 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x0000000002768800, is_vm_thread: true, completed in 0 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x000000000276e000, is_vm_thread: true, completed in 0 ns
[14.578s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x0000000017268800, is_vm_thread: true, completed in 0 ns
[14.579s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x000000001727e800, is_vm_thread: true, completed in 900 ns
# no log until 21.455s
[21.455s][6708][debug][handshake,task ] Operation: Deoptimize for thread 0x000000000286d800, is_vm_thread: true, completed in 12100 ns
[21.455s][6708][info ][handshake ] Handshake "Deoptimize", Targeted threads: 11, Executed by targeted threads: 0, Total completion time: 6876829000 ns
[21.455s][6708][debug][vmoperation ] end VM_Operation (0x000000000203d228): HandshakeAllThreads, mode: no safepoint, requested by thread 0x000000000286d800
[21.455s][7248][debug][class,resolve ] sun.security.jca.Providers java.lang.InheritableThreadLocal Providers.java:39
[21.455s][7248][info ][class,init ] 1251 Initializing 'java/lang/InheritableThreadLocal'(no method) (0x0000000100124740)
[21.455s][7248][debug][class,resolve ] java.lang.InheritableThreadLocal java.lang.ThreadLocal InheritableThreadLocal.java:57
[21.456s][7248][debug][class,preorder ] sun.security.jca.ProviderList source: C:\Users\example\AppData\Local\example\stable\jdk\lib\modules
[21.456s][7248][info ][class,load ] sun.security.jca.ProviderList source: jrt:/java.base
[21.456s][7248][debug][class,load ] klass: 0x00000001001249a8 super: 0x0000000100001080 loader: [loader data: 0x0000000002882bd0 of 'bootstrap'] bytes: 11522 checksum: bdc239d2
[21.456s][7248][debug][class,resolve ] sun.security.jca.ProviderList java.lang.Object (super)
...
The following two lines seems interesting:
[11.783s][6708][info ][handshake ] Handshake "Deoptimize", Targeted threads: 11, Executed by targeted threads: 0, Total completion time: 5536442500 ns
[21.455s][6708][info ][handshake ] Handshake "Deoptimize", Targeted threads: 11, Executed by targeted threads: 0, Total completion time: 6876829000 ns
Is that normal that these handshakes took 5.5 and 6.8 seconds?
I have experienced the same slowdown (and similar logs) with the update4j demo app (which is completely unrelated to our application) running with this command:
Z:\swing>\jdk-14\bin\java -Xlog:all=debug:file=app.txt:uptime,tid,level,tags:filecount=50 \
-jar update4j-1.4.5.jar --remote http://docs.update4j.org/demo/setup.xml
What should I look for to make our app faster again on single-CPU Windows 10 setups? Can I fix this by changing something in our application or by adding JVM arguments?
Is that a JDK bug, should I report it?
update 2020-04-25:
As far as I see the log files also contains GC logs. These are the first GC logs:
$ cat app.txt.00 | grep "\[gc"
[0.016s][7248][debug][gc,heap ] Minimum heap 8388608 Initial heap 60817408 Maximum heap 1073741824
[0.017s][7248][info ][gc,heap,coops ] Heap address: 0x00000000c0000000, size: 1024 MB, Compressed Oops mode: 32-bit
[0.018s][7248][info ][gc ] Using Serial
[22.863s][6708][info ][gc,start ] GC(0) Pause Young (Allocation Failure)
[22.863s][6708][debug][gc,heap ] GC(0) Heap before GC invocations=0 (full 0): def new generation total 17856K, used 15936K [0x00000000c0000000, 0x00000000c1350000, 0x00000000d5550000)
...
Unfortunately it does not seem related since it starts after the third pause.
update 2020-04-26:
With OpenJDK 14 the application uses 100% CPU in my (single-CPU) VirtualBox machine (running on a i7-6600U CPU). The virtual machine has 3,5 GB RAM. According to Task Manager 40%+ is free and disk activity is 0% (I guess this means no swapping). Adding another CPU to the virtual machine (and enabling hyper-threading for physical machines) make the application fast enough again. I just wondering, was it an intentional trade-off during JDK development to loss performance on (rare) single-CPU machines to make the JVM faster on multicore/hyper-threading CPUs?
TL;DR: It's an OpenJDK regression filed as JDK-8244340 and has been fixed in JDK 15 Build 24 (2020/5/20).
I did not except that but I could reproduce the issue with a simple hello world:
public class Main {
public static void main(String[] args) {
System.out.println("Hello world");
}
}
I have used these two batch files:
main-1cpu.bat, which limits the java process to only one CPU:
c:\windows\system32\cmd.exe /C start /affinity 1 \
\jdk-14\bin\java \
-Xlog:all=trace:file=app-1cpu.txt:uptime,tid,level,tags:filecount=50 \
Main
main-full.bat, the java process can use both CPUs:
c:\windows\system32\cmd.exe /C start /affinity FF \
\jdk-14\bin\java \
-Xlog:all=trace:file=app-full.txt:uptime,tid,level,tags:filecount=50 \
Main
(The differences are the affinity value and name of the log file. I've wrapped it for easier reading but wrapping with \ probably doesn't work on Windows.)
A few measurements on Windows 10 x64 in VirtualBox (with two CPUs):
PS Z:\main> Measure-Command { .\main-1cpu.bat }
...
TotalSeconds : 7.0203455
...
PS Z:\main> Measure-Command { .\main-full.bat }
...
TotalSeconds : 1.5751352
...
PS Z:\main> Measure-Command { .\main-full.bat }
...
TotalSeconds : 1.5585384
...
PS Z:\main> Measure-Command { .\main-1cpu.bat }
...
TotalSeconds : 23.6482685
...
The produced tracelogs contain similar pauses that you can see in the question.
Running Main without tracelogs is faster but the difference still can be seen between the single-CPU and two-CPU version: ~4-7 seconds vs. ~400 ms.
I've sent this findings to the hotspot-dev#openjdk mail list and devs there confirmed that this is something that the JDK could handle better. You can find supposed fixes in the thread too.
Another thread about the regression on hotspot-runtime-dev#. JIRA issue for the fix: JDK-8244340
From my experience performance problems with JDKs are related mostly to one of the following:
JIT Compilation
VM configuration (heap sizes)
GC algorithm
Changes in the JVM/JDK which break known good running applications
(Oh, and I forgot to mention class loading...)
If you just use the default JVM configuration since OpenJDK11, maybe you should set some of the more prominent options to fixed values, like GC, Heap size, etc.
Maybe some graphical analyse tool could help track your issue down. Like Retrace, AppDynamics or FlightRecorder and the like. These give more overview on the overall state of heap, GC cycles, RAM, threads, CPU load and so on at a given time than log files could provide.
Do I understand correctly that your application writes about 30710 lines to the log within the first second of running (under OpenJDK11)? Why is it "only" writing about 7k lines under OpenJDK14 in the first second? This seems like a huge difference for an application that is just started on different JVMs to me... Are you sure there are not for example high amounts of Exception stacktraces dumped into the log?
The other numbers are even higher sometimes, so maybe the slowdowns are related to exception logging? Or even swapping, if RAM gets low?
Actually I am thinking, if an application does not write anything into the log, this is a sign of smooth running without problems (unless it is frozen entirely in this time). What is happening from seconds 12-22 (in the OpenJDK14 case here) is what would concern me more... the logged lines go through the roof... why?
And afterwards the logging goes down to all time low values of about 1-2k lines... what is the reason for that?? (Well, maybe it is the GC kicking in at second 22 and does a tabula rasa which resolves some things...?)
Another thing may be your statement about "single CPU" machines. Does this imply "single core" also (Idk, maybe your software is tailored on legacy hardware or something)?
And the "single CPU" VMs are running on those machines?
But I assume, I am wrong about these assumptions, since almost all CPUs are multicore nowadays... but I would investigate on a multithreading issue (deadlock) problem maybe.
Since it's using 100% CPU "most of the time", and it takes 10 times longer (!) with Java 14, it means that you're wasting 90% of your CPU in Java 14.
Running out of heap space can do that, as you spend a whole lot of time in GC, but you seem to have ruled that out.
I notice that you're tweaking the biased locking option, and that it makes a significant difference. That tells me that maybe your program does a lot of concurrent work in multiple threads. It's possible that your program has a concurrency bug that shows up in Java 14, but not in Java 10. That could also explain why adding another CPU makes it more than twice as fast.
Concurrency bugs often only show up when you're unlucky, and the trigger could really have been anything, like a change to hashmap organization, etc.
First, if it's feasible, check for any loops that might be busy-waiting instead of sleeping.
Then, run a profiler in sampling mode (jvisualvm will do) and look for methods that are taking a much larger % of total time than they should. Since your performance is off by a factor of 10, any problems in there should really jump out.
This is an interesting issue and it would require indeterminate amount of effort to narrow it down since there are many permutations and combinations that need to be tried out and data collected and collated.
Seems as of there has been no resolution to this for some time. Perhaps this might need to be escalated.
EDIT 2: Since "ThreadLocalHandshakes" is deprecated and we can assume that locking is contended, suggest trying without "UseBiasedLocking" to hopefully speed up this scenario.
However there are some suggestions to collect more data and attempt to isolate the issue.
Allocate more than one core [I see that you have tried it and the issue goes away. Seems to be an issue with a thread/s execution precluding others. See no 7 below)
Allocate more heap (perhaps the demands of v14 is more than that of earlier jdks)
Allocate more memory to the Win 10 VB.
Check the OS system messages (Win 10 in your case)
Run it in an non-virtualized Win 10.
Try a different build of jdk 14
Do a thread dump every (or profile)few intervals of time. Analyze what thread is running exclusively. Perhaps there is a setting for equitable time sharing. Perhaps there is a higher priority thread running. What is that thread and what is it doing? In linux you could stat the lightweight processes (threads) associated with a process and its state in realtime. Something similar on Win 10?
CPU usage? 100% or less? Constrained by CPU or mem? 100% CPU in service threads? Which service thread?
Have you explicitly set a GC algo?
I have personally witnessed issues within versions that have to do with GC, heap resizing, issues with virtualized containers and so on.
There is no easy answer to that, I think, especially since this question has been around for some time. But we can try, all the best and let us know what is the result of some of these isolation steps.
EDIT 1: from the updated question, it seems to be related to a GC or another service thread taking over the single core non-equitably (Thread-Local Handshakes)?
Be careful with logging to slow disks, it will slow down your application:
https://engineering.linkedin.com/blog/2016/02/eliminating-large-jvm-gc-pauses-caused-by-background-io-traffic
But it doesn't seem likely to be the cause of the issue as the CPU is still busy and you don't have to wait for all threads to come to a safe point thanks to thread-local handshake: https://openjdk.java.net/jeps/312
Also not directly related to the problem you have but more generally if you want to try to squeeze more performance out of your hardware for startup time, take a look at AppCDS (class data sharing):
https://blog.codefx.org/java/application-class-data-sharing/
I'm developing a MS with Kotlin and Micronaut which access a Firestore database. When I run this MS locally I can make it work with 128M because it's very simple just read and write data to Firestore, and not big amounts of data, really small data like this:
{
"project": "DUMMY",
"columns": [
{
"name": "TODO",
"taskStatus": "TODO"
},
{
"name": "IN_PROGRESS",
"taskStatus": "IN_PROGRESS"
},
{
"name": "DONE",
"taskStatus": "DONE"
}
],
"tasks": {}
}
I'm running this in App Engine Standard in a F1 instance (256 MB 600 MHz) with this properties in my app.yaml
runtime: java11
instance_class: F1 # 256 MB 600 MHz
entrypoint: java -Xmx200m -jar MY_JAR.jar
service: data-connector
env_variables:
JAVA_TOOL_OPTIONS: "-Xmx230m"
GAE_MEMORY_MB: 128M
automatic_scaling:
max_instances: 1
max_idle_instances: 1
I know all that properties for handling memory are not necessary but I was desperate trying to make this work and just tried a lot of solutions because my first error message was:
Exceeded soft memory limit of 256 MB with 263 MB after servicing 1 requests total. Consider setting a larger instance class in app.yaml.
The error below is not fixed with the properties in the app.yaml, but now everytime I make a call to return that JSON I get this error
2020-04-10 12:09:15.953 CEST
While handling this request, the process that handled this request was found to be using too much memory and was terminated. This is likely to cause a new process to be used for the next request to your application. If you see this message frequently, you may have a memory leak in your application or may be using an instance with insufficient memory. Consider setting a larger instance class in app.yaml.
It always last longer in the first request, I think due to some Firestore configuration, but the thing is that I cannot make that work, always getting the same error.
Do you have any idea what I could be doing wrong or what I need to fix this?
TL;DR The problem was I tried to used a very small instance for a simple application, but even with that I needed more memory.
Ok, a friend helped me with this. I was using a very small instance and even when I didn't get the error of memory limit it was a memory problem.
Updating my instance to a F2 (512 MB 1.2 GHz) solved the problem and testing my app with siege resulted in a very nice performance:
Transactions: 5012 hits
Availability: 100.00 %
Elapsed time: 59.47 secs
Data transferred: 0.45 MB
Response time: 0.30 secs
Transaction rate: 84.28 trans/sec
Throughput: 0.01 MB/sec
Concurrency: 24.95
Successful transactions: 3946
Failed transactions: 0
Longest transaction: 1.08
Shortest transaction: 0.09
My sysops friends tells me that this instances are more for python scripting code and things like that, not JVM REST servers.