Java和ABAP的垃圾回收機制(Garbage Collection)比較

Both Java and ABAP can support automatic garbage collection. As an application developer the GC process are completely transparent for us and in our application code most of the time we should never call the GC API provided by Java or ABAP. Nevertheless it helps us to write more robust code with comparatively lower memory consumption if we have some basic understanding about the trigger point of Garbage collection.

Garbage Collection in Java

Use this small code below for test, which allocates 2MB memory every 100 millisecond.

Minor Garbage Collection in Java

Execute the program with the following VM argument.

The option PrintGCDetails can print out GC execution detail and Xmx40m means the max allowed heap allocation size is 40MB.

When executing you can find following output in console:

total :39424 KB

free:35720 KB

free:33672 KB

free:31624 KB

free:29576 KB

[GC (Allocation Failure) [PSYoungGen: 9847K->744K(11776K)] 9847K->752K(39424K), 0.0022270 secs] [Times: user=0.00 sys=0.00, real=0.00 secs]

free:36457 KB

free:34409 KB

free:32361 KB

free:30313 KB

[GC (Allocation Failure) [PSYoungGen: 9102K->616K(11776K)] 9110K->632K(39424K), 0.0018849 secs] [Times: user=0.00 sys=0.00, real=0.00 secs]

free:36564 KB

The GC log clearly shows that every time JVM fails to find a proper memory area in young generation, a GC operation is launched AUTOMATICALLY.

Some knowledge to understand this log

From Java official website we know the Heap in Hotspot( one kind of JVM ) consists of Young Generation, Old Generation and Permanent Generation.

( Reminder: Permanent Generation is removed from Java 8, Metaspace is introduced instead )

The Young Generation is where all new objects are allocated and aged. When the young generation fills up, this causes a minor garbage collection. This GC could be observed in our log that once GC is done, the free space size in young generation increases greatly.

All minor garbage collections are “Stop the World” events. This means that all application threads are stopped until the operation completes. The time spent on GC is also printed in the log, in our example:

Log:

total :1525248 KB

free:488140 KB

total :2549760 KB

free:488652 KB

total :3574272 KB

free:489164 KB

[GC (Allocation Failure) [PSYoungGen: 13107K->632K(152576K)] 3085107K->3072640K(4248576K), 0.0246054 secs] [Times: user=0.13 sys=0.00, real=0.02 secs]

[GC (Allocation Failure) [PSYoungGen: 632K->664K(283648K)] 3072640K->3072672K(4379648K), 0.0243200 secs] [Times: user=0.25 sys=0.00, real=0.03 secs]

[Full GC (Allocation Failure) [PSYoungGen: 664K->0K(283648K)] [ParOldGen: 3072008K->539K(268800K)] 3072672K->539K(552448K), [Metaspace: 2662K->2662K(1056768K)], 0.2419282 secs] [Times: user=0.11 sys=0.22, real=0.24 secs]

total :1576960 KB

free:547177 KB

total :2601472 KB

free:547689 KB

total :3625984 KB

free:548201 KB

[GC (Allocation Failure) [PSYoungGen: 5242K->0K(283648K)] 3077782K->3072539K(4379648K), 0.0127622 secs] [Times: user=0.03 sys=0.00, real=0.01 secs]

[GC (Allocation Failure) [PSYoungGen: 0K->0K(521728K)] 3072539K->3072539K(4617728K), 0.0115785 secs] [Times: user=0.13 sys=0.00, real=0.01 secs]

[Full GC (Allocation Failure) [PSYoungGen: 0K->0K(521728K)] [ParOldGen: 3072539K->1024K(275456K)] 3072539K->1024K(797184K), [Metaspace: 2662K->2662K(1056768K)], 0.2102163 secs] [Times: user=0.03 sys=0.17, real=0.21 secs]

total :1821696 KB

PARAMETERS: clear TYPE c as CHECKBOX DEFAULT abap_false.

TYPES: tt_table TYPE TABLE OF tadir WITH KEY pgmid object.
DATA: lt_result TYPE TABLE OF tadir,
      lt_total  TYPE TABLE OF tadir,
      lr_result TYPE REF TO tt_table.

DATA: c1 TYPE cursor.

OPEN CURSOR @c1 FOR SELECT * FROM tadir.

DO.
  WRITE: / |Round: { sy-index } | COLOR COL_NEGATIVE.
  CREATE DATA lr_result.
  FETCH NEXT CURSOR @c1 INTO TABLE @lr_result->* PACKAGE SIZE 800000.
  IF sy-subrc <> 0.
    EXIT.
  ENDIF.
  APPEND LINES OF lr_result->* TO lt_total.
  cl_abap_memory_utilities=>get_memory_size_of_object( EXPORTING object = lt_total
      IMPORTING
        bound_size_alloc = DATA(bound_alloc)
        bound_size_used = DATA(bound_used) ).

  WRITE: / 'bound alloc:' , bound_alloc.
  WRITE: / 'bound used:' , bound_used.

  cl_abap_memory_utilities=>get_total_used_size( IMPORTING size = DATA(lv_before_size) ).
  WRITE: / |Total size before GC: { lv_before_size }| COLOR COL_POSITIVE.
  IF clear = abap_true.
     CLEAR: lr_result->*, lt_result, lr_result.
  ENDIF.
  cl_abap_memory_utilities=>do_garbage_collection( ).
  cl_abap_memory_utilities=>get_total_used_size( IMPORTING size = DATA(lv_after_size) ).
  WRITE: / |Total size after GC: { lv_after_size }| COLOR COL_GROUP.
  DATA(rate) = ( lv_before_size - lv_after_size ) * 100 / lv_before_size.
  WRITE: / |Freed rate: { rate }%| COLOR COL_TOTAL.
ENDDO.

WRITE: / lines( lt_total ).
      

Compare the output of two execution we find out the final filled result, lt_total, are equal for both execution. And the temporary variable lr_result->*, lt_result and lr_result used to store the data queried from database table during each iteration are useless once the content are appended to the final result internal table, so in principle they are ok to be cleared manually. The comparison shows that if within a LOOP you have local variable which holds a large amount of data which is only handled in current loop, then you can manually clear it before the next round of loop starts to free up memory.

[外鍊圖檔轉存失敗,源站可能有防盜鍊機制,建議将圖檔儲存下來直接上傳(img-vehtBAFO-1598236863158)(https://upload-images.jianshu.io/upload_images/2085791-f48a0fb39770eaa8.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)]

You may argue that does the line below really take effect during the comparison?

cl_abap_memory_utilities=>do_garbage_collection( ).

Good question! If it is commented out and execute report with manual clear = abap_true, still the same output will be printed out as before it is commented out.

So let’s look at another example:

Execute it first with call_gc = abap_false, and the result shows even though you have manually clear the reference of lo_class, still no memory is free up.

[外鍊圖檔轉存失敗,源站可能有防盜鍊機制,建議将圖檔儲存下來直接上傳(img-OoCoEmUZ-1598236863159)(https://upload-images.jianshu.io/upload_images/2085791-3f0498690e66df0a.png?imageMogr2/auto-orient/strip%7CimageView2/2/w/1240)]

And the second time call_gc = true.

The GC processor has done a good job!

Last but not least, the purpose of this ABAP example is not asking you to call GC API manually in your application code, conversely, you should always avoid to do that.

Further reading

I have written a series of blogs which compare the language feature among ABAP, JavaScript and Java. You can find a list of them below:

Lazy Loading, Singleton and Bridge design pattern in JavaScript and in ABAP

Functional programming – Simulate Curry in ABAP

Functional Programming – Try Reduce in JavaScript and in ABAP

Simulate Mockito in ABAP

A simulation of Java Spring dependency injection annotation @Inject in ABAP

Singleton bypass – ABAP and Java

Weak reference in ABAP and Java

Java byte code and ABAP Load

How to write a “correct” program rejected by compiler: Exception handling in Java and in ABAP

An small example to learn Garbage collection in Java and in ABAP