Horst @spotcodereviews

Debugging, optimization, concurrency. Fixing code so you don't have to. Writing related articles. spotcodereviews.com Austria Joined March 2021

Tweets

50
Followers

75
Following

182
Likes

3

Ankur Kumar @ankurkumarz

5 years ago

It can't be better than this to visualize #Linux performance tools 👇 Courtesy: brendangregg.com/linuxperf.html

0 1 2 0 1

View Details

That's unfortunate. SecurityManager is still useful for dynamically restricting filesystem and network access for an application-specific scope. Also useful for dynamic low-overhead filesystem tracing. The only alternative that comes to mind is bytecode instrumentation.

OpenJDK @OpenJDK

5 years ago

New candidate JEP: 411: Deprecate the Security Manager for Removal: openjdk.java.net/jeps/411 #openjdk #java

9 52 132 0 4

0 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@tyvole @OpenJDK In that scenario I'd choose a whitelist approach at the very least. Sometimes, isolation at JVM granularity is too big of a cost. There are real world scenarios where filesystem accesses have to be disabled for a caller, for example. That's where SecurityManager can do its job.

0 0 1 0 0

View Details

Horst @spotcodereviews

5 years ago

@tyvole @OpenJDK With custom instrumentation, you have to be sure not to miss anything. For example, the tool does not seem to instrument AsynchronousFileChannel. Easy to fix, but that would be a security hole if the implementation was about security constraints. SecurityManager covers that.

1 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@tyvole @OpenJDK Sure, but it does use the alternative I mentioned: bytecode instrumentation :) More complex and definitely fine for tracing. For enforcing security constraints, a low-level facility would be better to cover all accesses.

1 0 1 0 0

View Details

Horst @spotcodereviews

5 years ago

@OpenJDK The only alternative I can think of is bytecode instrumentation.

1 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@OpenJDK SecurityManager is still useful for dynamically restricting filesystem and network access for an application-specific scope. It's also relevant for filesystem tracing if native facilities cannot be used or a low-overhead alternative is needed spotcodereviews.com/articles/debug…

1 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@seanjmullan SecurityManager is still useful for dynamically restricting filesystem and network access for an application-specific scope. Also useful for dynamic low-overhead filesystem tracing. The only alternative I can think of is bytecode instrumentation spotcodereviews.com/articles/debug…

0 0 0 0 0

View Details

Abandoned account of Thorsten Leemhuis (1/5) @kernellogger

5 years ago

All six parts from the "lockless patterns in #Linux" series @fagiolinux wrote for #LWNnet are now freely available: lwn.net/Articles/84422… "These articles will delve into the concepts behind #lockless algorithms and how they are used in the #kernel" Like this? Then subscribe!

0 51 169 0 40

View Details

Horst @spotcodereviews

5 years ago

A process terminated with a segmentation fault. There's no core dump. What can you do to get to the root cause? Here's one approach to save the day. spotcodereviews.com/articles/debug…

0 0 1 0 0

View Details

Roman Kennke 🗽 @rkennke

5 years ago

Project Lilliput is born 🐣 We've got a mailing list and project wiki now, code repository to appear soon. 😎openjdk.java.net/projects/lilli…

3 21 133 0 4

View Details

Horst @spotcodereviews

5 years ago

@SMT_Solvers @mwhudson @colinianking For memory analysis to be useful, the allocation site call stacks are essential. The full execution history is needed. The current/post-mortem process image lacks that info unless the allocator collected the call stacks (that's what the Valgrind tools Memcheck and Massif do).

0 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@SMT_Solvers @mwhudson @colinianking If I'm not mistaken, the screenshot in the article visualizes calls based on callgrind. For heap profiling or "who's responsible for allocating specific chunks of memory", massif (valgrind-based) would be my tool of choice. Can't be controlled like cachegrind, though.

0 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@mwhudson @SMT_Solvers @colinianking Depends on the tool. callgrind cannot be attached but can be started w/o instrumentation and controlled via callgrind_control --instr=<on|off> once an execution phase of interest is reached. The same can be done in code using so-called "client requests".

1 0 1 0 0

View Details

Horst @spotcodereviews

5 years ago

@__gparmer @samth @pcwalton @jorendorff Similar to increasing other limits (e.g. max memory mappings, max number of FDs). With a reasonably sized limit, you can catch undesired behavior (e.g. resource leaks) more quickly. Otherwise, you can get away with the issue for a longer period of time while consuming resources.

0 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@__gparmer @samth @pcwalton @jorendorff It's more of practical implication and applies to both implementation strategies. The net effect is a large stack with, if recursion is used without bounds, an excessive number of stack frames. Overflow happens when the upper limit is reached and that can "take a while".

1 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@__gparmer @samth @pcwalton @jorendorff Yes, it might kill another or the "very large stack" user during stack alloc. Reclamation would help, but with heavy stack usage fluctuations, you always have phases where large amounts are being committed. Not with a fully committed stack, but would also defeat the purpose :)

0 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@__gparmer @samth @pcwalton @jorendorff 3: If you actually use the large stack for recursion, you're potentially dealing with call stacks with thousands of frames - not practical for debugging/profiling, IMHO. Normally, that scenario only pops up while debugging an infinite recursion.

1 0 0 0 0

View Details

Horst @spotcodereviews

5 years ago

@__gparmer @samth @pcwalton @jorendorff 2: You're right. You can install guard pages, but your infinite recursion would have quite some room and therefore time to run, possible exhausting other resources while doing so. That would otherwise be caught sooner.