Show screenshot of per-process stats
authorSiraaj Khandkar <siraaj@khandkar.net>
Mon, 9 Dec 2019 16:18:47 +0000 (11:18 -0500)
committerSiraaj Khandkar <siraaj@khandkar.net>
Mon, 9 Dec 2019 16:18:47 +0000 (11:18 -0500)
README.md
screenshot--2015-10-05--18.41.30.jpg [new file with mode: 0644]

index 4402288..3709e2e 100644 (file)
--- a/README.md
+++ b/README.md
@@ -21,6 +21,20 @@ Essentially like `folsomite`, but different. Different in the following ways:
     + Well-defined metrics-to-binary conversions, as opposed to the
       nearly-arbitrary term-to-string conversions used in `folsomite`
     + Spec'd, tested and Dialyzed
+- More detailed stats:
+    - **per-process**. As much process ancestry is collected as possible, then
+      anonymous processes are aggregated to their youngest-known, named
+      predecessor - this aggregation keeps the useful breadcrumbs, while
+      reducing the number of unique names from exploding, which
+      **avoids the associated problems**:
+        1. not very useful when there're lots of short-lived processes
+        2. exploading disk space usage in Whisper
+    - per-ETS-table
+
+For an example of using pre-process stats to track-down memory leaks, here's a
+screenshot of the SSL connection process memory usage growth drop after upgrade
+from 17.5 to 18.1 (back in 2015):
+![SSL memory leak going away](screenshot--2015-10-05--18.41.30.jpg)
 
 ### Adding consumers
 
diff --git a/screenshot--2015-10-05--18.41.30.jpg b/screenshot--2015-10-05--18.41.30.jpg
new file mode 100644 (file)
index 0000000..b853720
Binary files /dev/null and b/screenshot--2015-10-05--18.41.30.jpg differ
This page took 0.023609 seconds and 4 git commands to generate.