maven-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <>
Subject [jira] [Commented] (WAGON-537) Maven download speed of large artifacts is slow due to unsuitable buffer strategy for remote Artifacts in AbstractWagon
Date Mon, 15 Oct 2018 10:17:00 GMT


ASF GitHub Bot commented on WAGON-537:

olaf-otto commented on a change in pull request #50: WAGON-537 Maven download speed of large
artifacts is slow

 File path: wagon-provider-api/src/main/java/org/apache/maven/wagon/
 @@ -560,31 +564,78 @@ protected void transfer( Resource resource, InputStream input, OutputStream
     protected void transfer( Resource resource, InputStream input, OutputStream output, int
requestType, long maxSize )
         throws IOException
-        byte[] buffer = new byte[DEFAULT_BUFFER_SIZE];
+        byte[] buffer = bufferForTransferring( resource );
         TransferEvent transferEvent = new TransferEvent( this, resource, TransferEvent.TRANSFER_PROGRESS,
requestType );
         transferEvent.setTimestamp( System.currentTimeMillis() );
         long remaining = maxSize;
         while ( remaining > 0 )
-            // let's safely cast to int because the min value will be lower than the buffer
-            int n = buffer, 0, (int) Math.min( buffer.length, remaining ) );
+            // Read from the stream, block if necessary until either EOF or buffer is filled.
+            // Filling the buffer has priority since downstream processors will significantly
degrade i/o
+            // performance if called to frequently (large data streams) as they perform expensive
tasks such as
+            // console output or data integrity checks.
+            int nextByte =;
-            if ( n == -1 )
+            if ( nextByte == -1 )
-            fireTransferProgress( transferEvent, buffer, n );
+            buffer[0] = ( byte ) nextByte;
+            // let's safely cast to int because the min value will be lower than the buffer
+            int length = (int) min( buffer.length, remaining ),
+                read = 1;
+            for ( ; read < length ; ++read )
+            {
+                nextByte =;
+                if ( nextByte == -1 )
+                {
+                    break;
+                }
+                buffer[read] = ( byte ) nextByte;
+            }
+            fireTransferProgress( transferEvent, buffer, read );
-            output.write( buffer, 0, n );
+            output.write( buffer, 0, read );
-            remaining -= n;
+            remaining -= read;
+    /**
+     * Provide a buffer suitably sized for efficiently
+     * {@link #transfer(Resource, InputStream, OutputStream, int, long) transferring}
+     * the given {@link Resource}. For larger files, larger buffers are provided such that
+     * {@link #fireTransferProgress(TransferEvent, byte[], int) listeners} are not notified
overly frequently.
+     * For instance, transferring gigabyte-sized resources would result in millions of notifications
when using
+     * only a few kilobytes of buffer, drastically slowing transfer since transfer progress
listeners and
+     * notifications are synchronous and may block, e.g. when writing download progress status
to console.
+     *
+     * @param resource must not be null.
+     * @return a byte buffer suitable for the {@link Resource#getContentLength() content
length} of the resource.
+     */
+    protected byte[] bufferForTransferring( Resource resource )
+    {
+        long contentLength = resource.getContentLength();
+        if ( contentLength <= 0 )
+        {
+            return new byte[DEFAULT_BUFFER_SIZE];
+        }
+        int numberOf4KbSegmentsFor100Chunks = ( ( int ) ( contentLength / ( 4 * 1024 * 100
) ) );
 Review comment:
   Kindly asking to specifically review the way the buffer is sized here. 
   The Idea is that at least the old buffer size (DEFAULT_BUFFER_SIZE) is used and that buffers
should have a size of N * 4Kb. As a rule of thumb, the buffer should be sized such that at
least 100 chunks of data are being processed. I've capped the buffer at MAXIMUM_BUFFER_SIZE
(512 Kb).
   Resulting, most of the files transferred (Metadata such as POMs and checksums, small JARs...)
are transferred with the old buffer size. However, larger files are transferred with up to
512 Kb of buffer.
   For a 10GB stream, this will reduce the number of chunks from > 5 million (the previous
strategy almost never used the entire 4k buffer but less than half) to about 20 thousand chunks.

This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:

> Maven download speed of large artifacts is slow due to unsuitable buffer strategy for
remote Artifacts in AbstractWagon
> -----------------------------------------------------------------------------------------------------------------------
>                 Key: WAGON-537
>                 URL:
>             Project: Maven Wagon
>          Issue Type: Improvement
>          Components: wagon-provider-api
>    Affects Versions: 3.2.0
>         Environment: Windows 10, JDK 1.8, Nexus  Artifact store > 100MB/s network
>            Reporter: Olaf Otto
>            Priority: Major
>              Labels: perfomance
>         Attachments: wagon-issue.png
> We are using maven for build process automation with docker. This sometimes involves
downloading images with a few gigabytes in size. Here, maven's download speed is consistently
and reproducibly slow. For instance, an artifact with 7,5 GB in size took almost two hours
to transfer in spite of a 100 MB/s connection with respective reproducible download speed
from the remote nexus artifact repository when using a browser to download.
> I have investigated the issue using JProfiler. The result clearly shows a significant
issue in AbstractWagon's transfer( Resource resource, InputStream input, OutputStream output,
int requestType, long maxSize ) method used for remote artifacts.
> Here, the input stream is read in a loop using a 4 Kb buffer. Whenever data is received,
the received data is pushed to downstream listeners via fireTransferProgress. These listeners
(or rather consumers) perform  expensive tasks such as checksumming or printing to console.
> Now, the underlying InputStream implementation used in transfer will return calls to
read(bugger, offset, length) as soon as *some* data is available. That is, fireTransferProgress
is invoked with an average number of bytes less than half the buffer capacity (this varies
with the underlying network and hardware architecture). Consequently, fireTransferProgress
is invoked *millions of times* for large files. As this is a blocking operation, the time
spent in fireTransferProgress dominates and drastically slows down the transfer by at least
one order of magnitude. 
> !wagon-issue.png! 
> In our case, we found download speed reduced from a theoretical optimum of ~80 seconds
to to more than 3200 seconds.
> From an architectural perspective, I would not want to make the consumers / listeners
invoked via fireTransferProgress aware of their potential impact on download speed, but rather
refactor the transfer method such that it uses a buffer strategy reducing the the number of
fireTransferProgress invocations. This should be done with regard to the expected file size
of the transfer, such that fireTransferProgress is invoked often enough but not to frequent.
> I have implemented a solution and transfer speed went up more than one order of magnitude.
I will provide a pull request asap.

This message was sent by Atlassian JIRA

View raw message