RFR: 8257234 : Add gz option to SA jmap to write a gzipped heap dump [v14]

[Serguei Spitsyn]

On Thu, 4 Feb 2021 23:23:41 GMT, Chris Plummer <cjplummer at openjdk.org> wrote:

>> Thanks for the thorough list, I have tested most of them but not all. I will cover them and update here later.
>> 
>>> So really there are two implementations of heap dumping, one in the VM and one in SA, but a total of 5 ways to get to them. I'm just starting to get my head wrapped around all this myself. We've created a very confusing situation for us, and unfortunately I think some of the changes that got us here was driven by a goal to simply things.
>> 
>> Yes, it makes me even curious about why there can be two different implementation of them (one by C++ and one by Java),  it seems we need double the work of adding new options for those tools, like parallal for jmap histo and dump, etc.
>
> One implementation is in the JVM itself, to be used when the JVM is still running well, and not just from command line tools. Heap dumping can also be triggered by the JVM itslef by setting flags like -XX:+HeapDumpBeforeFullGC. The other implementation is in SA, to be used on a core file or a hung JVM process, although it can also be used on a JVM that is still running well. BTW, there used to be a 3rd implementation. The old hprof profiler (a JVMTI agent) also was used to generate hprof files.

Hi Lin,

A week ago you replied that you are accepting the following suggested refactoring:

                   /*
                    * Possible commands:
                    *     dumpheap gz=1 file
                    *     dumpheap gz=1
                    *     dumpheap file
                    *     dumpheap
                    *
                    * Use default filename if cntTokens == 0.
                    * Handle cases with cntTokens == 1 or 2.
                    */

                    if (cntTokens > 2) {
                        err.println("Too big number of options: " + cntTokens);
                        usage();
                        return;
                    }
                    if (cntTokens >= 1) { // parse first argument which is "gz=" option
                        String option  = t.nextToken();
                        gzlevel = parseHeapDumpCompressionLevel(option);
                        if (gzlevel == 0) {
                            usage();
                            return;
                        }
                        filename = "heap.bin.gz";
                    }
                    if (cntTokens == 2) { // parse second argument which is filename
                        filename = t.nextToken();
                        if (filename.startsWith("gz=")) {
                            err.println("Filename should not start with "gz=": " + filename);
                            usage();
                            return;
                        }
                    }

But I still do not see it in the latest updates from you. It feels like there is some misunderstanding and confusion.
Could you, please, clear this up?

Thanks,
Serguei

-------------

PR: https://git.openjdk.java.net/jdk/pull/1712