[CAMEL-21288] Log processor causing memory leak in split for very large data sets

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 4.1.0
• Fix Version/s: 4.2.0
• Component/s: camel-core
• Estimated Complexity: Unknown
• Assignee: Unassigned
• Reporter: Michal Stepan
• Created: 30/Sep/24 08:09
• Updated: 03/Oct/24 06:46
• Resolved: 03/Oct/24 06:46

Description

Given random data generator function:

 

public static List<Map<String, Object>> seed(int numberOfRows, int numberOfColumns) {
    List<Map<String, Object>> dataList = new ArrayList<>();
    Random random = new Random();
    for (int i = 0; i < numberOfRows; i++) {
        Map<String, Object> row = new HashMap<>();
        for (int j = 1; j <= numberOfColumns; j++) {
            String columnName = "col" + j;
            var value = random.nextInt(1000);
            row.put(columnName, value);
        }
        dataList.add(row);
    }
    return dataList;
} 

And two processors - first generates 20 batches and second would generate 20k rows in each batch (tweak as you want):

 

public class OutsideSplitProcessor implements Processor {

    @Override
    public void process(Exchange exchange) throws Exception {
        exchange.getIn().setBody(seed(20, 1));
    }
} 

 

public class InsideSplitProcessor implements Processor {
    
    @Override
    public void process(Exchange exchange) throws Exception {
        exchange.getIn().setBody(seed(20000, 20));
    }
} 

And a route:

 

<route>
 <from uri="direct:test"/>
 <process ref="outsideSplitProcessor"/> 
 <split stopOnException="true" parallelProcessing="false" streaming="true">
  <simple>${body}</simple>
  <process ref="insideSplitProcessor"/>
  <log message="Ha, now you fail ${body.size()}"/>
  <setBody><constant/></setBody>
 </split>
 <to uri="mock:test"/>
</route> 

The processing would fail on OOM when used limited memory setting ( -Xmx512m in my case of macbook m1 pro 16Gb ram).

The problem is on the line:

 

<log message="Ha, now you fail ${body.size()}"/> 

Where upon analysis, the expression evaluation stores the content of the body into memory (ok), but keep it referrenced even after leaving the split. This is happening only when the generated data are objects (Random usage in this case) - when using unboxed int values, the problem is not there. Our original case was using sql component, that returned database data (boxed in objects).

 

You can mitigate the problem by using external processor instead of log:

 

<process ref="logProcessor"/> 

public class LogProcessor implements Processor {

    @Override
    public void process(Exchange exchange) throws Exception {
        log.info("Haha, now you will not fail: {}", exchange.getIn().getBody(List.class).size());
    }
} 

or using groovy:

<groovy>
    request.headers.bodySize = body.size()
</groovy> 

In both cases, referrences are cleaned up - not causing OOM.

 

This behavior seems very unexpected.