Additional heap dump features

SapMachine includes some additional VM flags related to creating heap dumps.

Allow overwriting the heap dump file created by -XX:+HeapDumpOnOutOfMemoryError

By default, when a heap dump is created via -XX:+HeapDumpOnOutOfMemoryError, the dumping fails when the file already exists. This can be avoided with the -XX:+HeapDumpOverwrite VM flag.

Allow dumping the heap to a special file like a tty or a named pipe

On a disk space limited system, dumping the heap to a tty or named pipe can be used to stream the heap dump to another system, without needing local disk space.

E.g. if you can connect to a system via ssh, you can write the heap dump to the tty device of the ssh connection and redirect the ssh output to a file on the connecting machine.

Similarly, you could create a named pipe via mkfifo and dump the heap to this file. Then a program like nc can be used to read from the named pipe and stream it via the network to another server. Note that named pipes on Windows are not regular files and thus cannot be used as the destination of a heap dump.

Since these types of files already exist, the -XX:+HeapDumpOverwrite flag must be set.

Starting from JDK 25, the heap dump is not written as a single file anymore. This would defeat the purpose of writing to special files. It can be avoided via the -XX:HeapDumpParallelism=1 VM flag. This disables parallel heap dump writing and avoids using more than the specified file.

Create smaller compressed heap dump files

Heap dumps can already be created in compressed gzip format (via -XX:HeapDumpGzipLevel=1 for dumps created by -XX:+HeapDumpOnOutOfMemoryError or with the -gz option for the GC.heap_dump jcmd).

If the dump is too large even after compression, the -XX:+LimitPrimitiveArrayContentInHeapDump can be used to further decrease the size of the compressed heap dump. This is done by omitting the content of large primitive arrays. More specifically, only the first ‘N’ elements of these arrays contain the real values and the later values are replaced by 0 or false (for boolean arrays). This makes compression of these arrays more efficient.

By default, only the first 120 elements of char and byte arrays are preserved with -XX:+LimitPrimitiveArrayContentInHeapDump and the first 50 elements of the other primitive arrays. The larger value for char and byte is used since these arrays often contain string content we want to preserve (e.g. class names).

These default values can be changed with -XX:StringLikeContentSizeLimitInHeapDump=<prefix-size> for char and byte arrays and with -XX:ArrayContentSizeLimitInHeapDump=<prefix-size> for the other primitive array types.