feat: sub-interpreters support

ChangeLog

@@ -1,5 +1,7 @@
 2023-xx-xx v3.6.0
 
+  Added support for sub-interpreters.
+
   Dropped support for Python 2, 3.3, 3.4, 3.5, 3.6 and 3.7.
 
   Bugfix: ensure that threads are resumed by austinp when an error occurs during

README.md

@@ -367,11 +367,12 @@ for some further processing.
 By default, each line has the following structure:
 
 ~~~
-P<pid>;T<tid>[;[frame]]* [metric]*
+P<pid>;T<iid>:<tid>[;[frame]]* [metric]*
 ~~~
 
 where the structure of `[frame]` and the number and type of metrics on each line
-depend on the mode.
+depend on the mode. The `<pid>`, `<iid>` and `<tid>` component represent the
+process ID, the sub-interpreter ID, and the thread ID respectively.
 
 
 ## Environment variables
@@ -449,6 +450,15 @@ Austin can be told to profile multi-process applications with the `-C` or
 process.
 
 
+## Sub-interpreters
+
+Austin has support for Python applications that make use of sub-interpreters.
+This means that Austin will sample all the sub-interpreters that are running
+within each process making up the Python application.
+
+*Since Austin 3.6.0*.
+
+
 ## Garbage Collector Sampling
 
 Austin can sample the Python garbage collector state for applications running

scripts/requirements-bm.txt

@@ -1,2 +1,2 @@
-austin-python~=1.4.1
+austin-python~=1.6
 scipy~=1.10.1

scripts/requirements-val.txt

@@ -1,3 +1,3 @@
-austin-python~=1.5
+austin-python~=1.6
 numpy
 scipy

src/argparse.c

@@ -56,11 +56,11 @@ const char SAMPLE_FORMAT_KERNEL[]      = ";kernel:%s:0";
 const char SAMPLE_FORMAT_WHERE_KERNEL[]= "    \033[38;5;159m%s\033[0m 🐧\n";
 #endif
 #if defined PL_WIN
-const char HEAD_FORMAT_DEFAULT[]       = "P%I64d;T%I64x";
-const char HEAD_FORMAT_WHERE[]         = "\n\n%3$s%4$s Process \033[35;1m%1$I64d\033[0m 🧵 Thread \033[34;1m%2$I64d\033[0m\n\n";
+const char HEAD_FORMAT_DEFAULT[]       = "P%I64d;T%I64x:%I64x";
+const char HEAD_FORMAT_WHERE[]         = "\n\n%4$s%5$s Process \033[35;1m%1$I64d\033[0m 🧵 Thread \033[34;1m%2$I64d:%3$I64d\033[0m\n\n";
 #else
-const char HEAD_FORMAT_DEFAULT[]       = "P%d;T%ld";
-const char HEAD_FORMAT_WHERE[]         = "\n\n%3$s%4$s Process \033[35;1m%1$d\033[0m 🧵 Thread \033[34;1m%2$ld\033[0m\n\n";
+const char HEAD_FORMAT_DEFAULT[]       = "P%d;T%ld:%ld";
+const char HEAD_FORMAT_WHERE[]         = "\n\n%4$s%5$s Process \033[35;1m%1$d\033[0m 🧵 Thread \033[34;1m%2$ld:%3$ld\033[0m\n\n";
 #endif
 
 

src/events.h

@@ -75,13 +75,13 @@
     }                                      \
   }
 
-#define emit_stack(format, pid, tid, ...)                         \
-  {                                                               \
-    if (pargs.binary) {                                           \
-      mojo_stack(pid, tid);                                       \
-    } else {                                                      \
-      fprintfp(pargs.output_file, format, pid, tid, __VA_ARGS__); \
-    }                                                             \
+#define emit_stack(format, pid, iid, tid, ...)                         \
+  {                                                                    \
+    if (pargs.binary) {                                                \
+      mojo_stack(pid, iid, tid);                                       \
+    } else {                                                           \
+      fprintfp(pargs.output_file, format, pid, iid, tid, __VA_ARGS__); \
+    }                                                                  \
   }
 
 #define emit_frame_ref(format, frame)                                      \

src/mojo.h

@@ -29,7 +29,7 @@
 #include "cache.h"
 #include "platform.h"
 
-#define MOJO_VERSION 2
+#define MOJO_VERSION 3
 
 enum {
   MOJO_RESERVED,
@@ -119,9 +119,10 @@ static inline void mojo_integer(mojo_int_t integer, int sign) {
   mojo_string(label);             \
   mojo_fstring(__VA_ARGS__);
 
-#define mojo_stack(pid, tid) \
-  mojo_event(MOJO_STACK);    \
-  mojo_integer(pid, 0);      \
+#define mojo_stack(pid, iid, tid) \
+  mojo_event(MOJO_STACK);         \
+  mojo_integer(pid, 0);           \
+  mojo_integer(iid, 0);           \
   mojo_fstring(FORMAT_TID, tid);
 
 #define mojo_frame(frame)           \

src/py_proc.c

@@ -1166,6 +1166,7 @@ _py_proc__sample_interpreter(py_proc_t * self, PyInterpreterState * is, ctime_t
       current_thread = _py_proc__get_current_thread_state_raddr(self);
   }
 
+  int64_t interp_id = V_FIELD_PTR(int64_t, is, py_is, o_id);
   do {
     if (pargs.memory) {
       mem_delta = 0;
@@ -1183,6 +1184,7 @@ _py_proc__sample_interpreter(py_proc_t * self, PyInterpreterState * is, ctime_t
 
     py_thread__emit_collapsed_stack(
       &py_thread,
+      interp_id,
       time_delta,
       mem_delta
     );
@@ -1200,45 +1202,54 @@ _py_proc__sample_interpreter(py_proc_t * self, PyInterpreterState * is, ctime_t
 // ----------------------------------------------------------------------------
 int
 py_proc__sample(py_proc_t * self) {
-  ctime_t time_delta = gettime() - self->timestamp;  // Time delta since last sample.
+  ctime_t   time_delta     = gettime() - self->timestamp;  // Time delta since last sample.
+  void    * current_interp = self->is_raddr;
 
   V_DESC(self->py_v);
 
   PyInterpreterState is;
-  if (fail(py_proc__get_type(self, self->is_raddr, is))) {
-    log_ie("Failed to get interpreter state while sampling");
-    FAIL;
-  }
 
-  void * tstate_head = V_FIELD(void *, is, py_is, o_tstate_head);
-  if (!isvalid(tstate_head))
-    // Maybe the interpreter state is in an invalid state. We'll try again
-    // unless there is a fatal error.
-    SUCCESS;
+  do {
+    if (fail(py_proc__get_type(self, current_interp, is))) {
+      log_ie("Failed to get interpreter state while sampling");
+      FAIL;
+    }
 
-  #ifdef NATIVE
-  raddr_t raddr = { .pref = self->proc_ref, .addr = tstate_head };
-  if (fail(_py_proc__interrupt_threads(self, &raddr))) {
-    log_ie("Failed to interrupt threads");
-    FAIL;
-  }
-  time_delta = gettime() - self->timestamp;
-  #endif
+    void * tstate_head = V_FIELD(void *, is, py_is, o_tstate_head);
+    if (!isvalid(tstate_head))
+      // Maybe the interpreter state is in an invalid state. We'll try again
+      // unless there is a fatal error.
+      SUCCESS;
 
-  int result = _py_proc__sample_interpreter(self, &is, time_delta);
+    #ifdef NATIVE
+    raddr_t raddr = { .pref = self->proc_ref, .addr = tstate_head };
+    if (fail(_py_proc__interrupt_threads(self, &raddr))) {
+      log_ie("Failed to interrupt threads");
+      FAIL;
+    }
+    time_delta = gettime() - self->timestamp;
+    #endif
+
+    int result = _py_proc__sample_interpreter(self, &is, time_delta);
 
+    #ifdef NATIVE
+    if (fail(_py_proc__resume_threads(self, &raddr))) {
+      log_ie("Failed to resume threads");
+      FAIL;
+    }
+    #endif
+
+    if (fail(result))
+      FAIL;
+  } while (isvalid(current_interp = V_FIELD(void *, is, py_is, o_next)));
+
   #ifdef NATIVE
   self->timestamp = gettime();
-
-  if (fail(_py_proc__resume_threads(self, &raddr))) {
-    log_ie("Failed to resume threads");
-    FAIL;
-  }
   #else
   self->timestamp += time_delta;
   #endif
 
-  return result;
+  SUCCESS;
 } /* py_proc__sample */
 
 

src/py_thread.c

@@ -893,7 +893,7 @@ py_thread__next(py_thread_t * self) {
 
 // ----------------------------------------------------------------------------
 void
-py_thread__emit_collapsed_stack(py_thread_t * self, ctime_t time_delta, ssize_t mem_delta) {
+py_thread__emit_collapsed_stack(py_thread_t * self, int64_t interp_id, ctime_t time_delta, ssize_t mem_delta) {
   if (!pargs.full && pargs.memory && mem_delta == 0)
     return;
 
@@ -924,7 +924,7 @@ py_thread__emit_collapsed_stack(py_thread_t * self, ctime_t time_delta, ssize_t
 
   // Group entries by thread.
   emit_stack(
-    pargs.head_format, self->proc->pid, self->tid,
+    pargs.head_format, self->proc->pid, interp_id, self->tid,
     // These are relevant only in `where` mode
     is_idle           ? "💤" : "🚀",
     self->proc->child ? "🧒" : ""

src/py_thread.h

@@ -92,11 +92,12 @@ py_thread__next(py_thread_t *);
  * Print the frame stack using the collapsed format.
  *
  * @param  py_thread_t  self.
+ * @param  int64_t      the interpreter ID.
  * @param  ctime_t      the time delta.
  * @param  ssize_t      the memory delta.
  */
 void
-py_thread__emit_collapsed_stack(py_thread_t *, ctime_t, ssize_t);
+py_thread__emit_collapsed_stack(py_thread_t *, int64_t, ctime_t, ssize_t);
 
 
 /**

src/python/interp.h

@@ -41,6 +41,7 @@ struct _ts; /* Forward */
 typedef struct _is2 {
     struct _is2 *next;
     struct _ts *tstate_head;
+    int64_t id;
     void* gc;  /* Dummy */
 } PyInterpreterState2;
 

src/version.h

@@ -163,6 +163,7 @@ typedef struct {
 
   offset_t o_next;
   offset_t o_tstate_head;
+  offset_t o_id;
   offset_t o_gc;
   offset_t o_gil_state;
 } py_is_v;
@@ -303,20 +304,23 @@ typedef struct {
   sizeof(s),                            \
   offsetof(s, next),                    \
   offsetof(s, tstate_head),             \
+  offsetof(s, id),                      \
   offsetof(s, gc),                      \
 }
 
 #define PY_IS_311(s) {                  \
   sizeof(s),                            \
   offsetof(s, next),                    \
   offsetof(s, threads.head),            \
+  offsetof(s, id),                      \
   offsetof(s, gc),                      \
 }
 
 #define PY_IS_312(s) {                  \
   sizeof(s),                            \
   offsetof(s, next),                    \
   offsetof(s, threads.head),            \
+  offsetof(s, id),                      \
   offsetof(s, gc),                      \
   offsetof(s, ceval.gil),               \
 }

test/functional/test_fork.py

@@ -53,6 +53,7 @@ def test_fork_wall_time(austin, py, heap, mojo):
     assert len(processes(result.stdout)) == 1, compress(result.stdout)
     ts = threads(result.stdout)
     assert len(ts) == 2, compress(result.stdout)
+    assert all(len(t[1].split(":")) == 2 for t in ts), "threads have interpreter ID"
 
     assert has_pattern(result.stdout, "target34.py:keep_cpu_busy:3"), compress(
         result.stdout

test/requirements.txt

@@ -1,4 +1,4 @@
-austin-python~=1.5
+austin-python~=1.6
 flaky
 pytest
 pytest-xdist