Inverted flame graph #439

ivonastojanovic · 2023-08-14T10:12:19Z

Added inverted flame graph which is generated with --inverted command line argument. The tree's nodes with the same parent and location are merged when we check 'Hide Import System Frames' checkbox.

Fixes: #293

Describe your changes
A clear and concise description of the changes you have made.

Testing performed
Describe the testing you have performed to ensure that the bug has been addressed, or that the new feature works as planned.

Additional context
Add any other context about your contribution here.

pablogsal · 2023-08-14T12:03:25Z

src/memray/reporters/flamegraph.py

@@ -75,6 +75,21 @@ def __init__(
        self.data = data
        self.memory_records = memory_records

+    @classmethod
+    def update_import_system_frames_for_inverted_flamegraph(


This may have two problems:

For deep trees, recursion can be quite slow.

In Python there is a recursion limit to protect from stack overflows. This means that if the tree is deep enough this may blow up with a RecursionError.

We could transform this function into something iterative that keeps the stack manually but on the other hand we are already limiting the stack depth here:

memray/src/memray/reporters/flamegraph.py

Lines 164 to 167 in 0154228

if index - num_skipped_frames > MAX_STACKS:

current_frame["name"] = "<STACK TOO DEEP>"

current_frame["location"] = ["...", "...", 0]

break

It may be good to add a test to see that we don't crash for a stack that's too deep. The test would just see what's the generated data for an input that's very nested.

pablogsal · 2023-08-14T12:05:00Z

src/memray/reporters/flamegraph.py

@@ -178,6 +195,11 @@ def _from_any_snapshot(
                    for interval in intervals
                )

+        if inverted:
+            frames[0][


Is the assignment necessary? update_import_system_frames_for_inverted_flamegraph does this:

nodes[node_index]["import_system"] |= result

and on the first call node_index is 0 so it should update frames[0] already, no?

pablogsal · 2023-08-14T12:06:19Z

src/memray/reporters/templates/__init__.py

@@ -28,7 +28,9 @@ def include_file(name: str) -> Markup:
    return env


-def get_report_title(*, kind: str, show_memory_leaks: bool) -> str:
+def get_report_title(*, kind: str, show_memory_leaks: bool, inverted: bool) -> str:


I think you missed some updates of calls to this function for example here: tests/unit/test_templates.py

pablogsal · 2023-08-14T12:30:20Z

src/memray/reporters/assets/common.js

+      node.value += same_node.value;
+    }
+
+    root.children = root.children.filter(


Hummmm. isn't this invalidating the array root.children and therefore making the indexing incorrect?

pablogsal · 2023-08-14T12:31:19Z

src/memray/reporters/assets/common.js

@@ -192,6 +192,32 @@ function filterFramesByFunc(root, func) {
  return _.defaults({ children: children }, root);
 }

+function mergeNodes(root) {


This function will likely need some tests in src/memray/reporters/assets/common.test.js

codecov-commenter · 2023-08-21T13:21:39Z

Codecov Report

Patch coverage: 91.37% and project coverage change: +0.16% 🎉

Comparison is base (0154228) 91.72% compared to head (9037cba) 91.89%.
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #439      +/-   ##
==========================================
+ Coverage   91.72%   91.89%   +0.16%     
==========================================
  Files          90       90              
  Lines       10532    10766     +234     
  Branches     1450     1472      +22     
==========================================
+ Hits         9661     9893     +232     
- Misses        868      871       +3     
+ Partials        3        2       -1

Flag	Coverage Δ
cpp	`85.23% <ø> (+0.02%)`	⬆️
python_and_cython	`95.30% <91.37%> (+0.11%)`	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files Changed	Coverage Δ
tests/unit/test_flamegraph_reporter.py	`100.00% <ø> (ø)`
src/memray/reporters/table.py	`94.28% <50.00%> (-2.69%)`	⬇️
src/memray/reporters/transform.py	`96.29% <50.00%> (-1.79%)`	⬇️
src/memray/commands/common.py	`86.53% <80.00%> (-0.34%)`	⬇️
src/memray/reporters/flamegraph.py	`91.66% <93.20%> (+3.66%)`	⬆️
src/memray/_ipython/flamegraph.py	`95.45% <100.00%> (+0.06%)`	⬆️
src/memray/commands/flamegraph.py	`100.00% <100.00%> (ø)`
tests/unit/test_templates.py	`100.00% <100.00%> (ø)`

... and 1 file with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

godlygeek · 2023-08-21T19:37:50Z

src/memray/reporters/flamegraph.py

 from typing import TextIO
 from typing import Tuple
+from typing import TypedDict


CI is failing because of this. We still support 3.7, and typing.TypedDict wasn't added until Python 3.8. We either need to use from typing_extensions import TypedDict on Python 3.7, or drop support for Python 3.7 - or maybe make this conditional on Python version... or maybe only use TypedDict when typing.TYPE_CHECKING is set.

Or maybe just use a dict for now, not worry about typing it since it's an internal API, and add an issue to remind us to add typing for it after we drop 3.7 support 😄

godlygeek · 2023-08-21T19:39:17Z

news/439.feature.rst

@@ -0,0 +1,4 @@
+Add a new ``inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,


Suggested change

Add a new ``inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,

Add a new ``--inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,

godlygeek · 2023-08-21T19:39:31Z

news/439.feature.rst

@@ -0,0 +1,4 @@
+Add a new ``inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,
+and the children of any given node represent the percentage of any of that node's


Suggested change

and the children of any given node represent the percentage of any of that node's

and the children of any given node represent the percentage of that node's

godlygeek · 2023-08-21T19:39:49Z

news/439.feature.rst

@@ -0,0 +1,4 @@
+Add a new ``inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,
+and the children of any given node represent the percentage of any of that node's
+allocations that can be attributed to a particular caller.The inverted flame graph is very helpful in analyzing where memory is being


Suggested change

allocations that can be attributed to a particular caller.The inverted flame graph is very helpful in analyzing where memory is being

allocations that can be attributed to a particular caller. The inverted flame graph is very helpful in analyzing where memory is being

godlygeek · 2023-08-21T19:42:29Z

src/memray/_ipython/flamegraph.py

@@ -90,6 +90,13 @@ def argument_parser() -> argparse.ArgumentParser:
        default=False,
    )

+    parser.add_argument(
+        "--inverted",
+        help="Invert flame graph",


We should explain this in more detail, I think. Something like:

Suggested change

help="Invert flame graph",

help=(

"Invert the flame graph: "

"use allocators as roots instead of thread entry points",

)

godlygeek · 2023-08-21T21:02:53Z

src/memray/reporters/flamegraph.py

@@ -132,14 +178,15 @@ def _from_any_snapshot(
            )
            num_skipped_frames = 0
            is_import_system = False
-            for index, stack_frame in enumerate(reversed(stack)):
+            stack_it = enumerate((stack)) if inverted else enumerate(reversed(stack))


Suggested change

stack_it = enumerate((stack)) if inverted else enumerate(reversed(stack))

stack_it = enumerate(stack) if inverted else enumerate(reversed(stack))

godlygeek · 2023-08-21T22:07:44Z

src/memray/reporters/flamegraph.py

+            result = cls.update_import_system_frames_for_inverted_flamegraph(
+                nodes, child_index
+            )
+            nodes[node_index]["import_system"] |= result


The |= here is essentially saying: if any of this node's callers are part of the import system, then this is an import system node. That doesn't seem correct to me. Consider a test case like this:

def test_inverted_allocator_sometimes_called_under_importlib(self): # GIVEN peak_allocations = [ MockAllocationRecord( tid=1, address=0x1000000, size=1024, allocator=AllocatorType.MALLOC, stack_id=1, n_allocations=1, _stack=[ ("malloc", "malloc.c", 12), ("parent_one", "fun.py", 8), ("main", "main.py", 4), ], ), MockAllocationRecord( tid=1, address=0x1000000, size=1024, allocator=AllocatorType.MALLOC, stack_id=1, n_allocations=1, _stack=[ ("malloc", "malloc.c", 12), ("importer", "<frozen importlib>", 4), ("parent_two", "fun.py", 10), ("main", "main.py", 4), ], ), ]

In that example, malloc can be called from two different paths - either by main calling parent_one calling malloc, or by main calling parent_two calling importer calling malloc. importer is part of the import system, so there's one path to malloc that goes through the import system and one that doesn't. Each of those paths are responsible for half the bytes allocated by malloc.

Our non-inverted flame graph displays these stacks for that:

| root size=2048 | | main size=2048 | | parent_one size=1024 | parent_two size=1024 | | malloc size=1024 | importer size=1024 | | | malloc size=1024 |

When you click the "hide import system frames" checkbox, importer and the frames below it are hidden, and parent_two becomes a leaf:

| root size=2048 | | main size=2048 | | parent_one size=1024 | parent_two size=1024 | | malloc size=1024 | |

For our inverted flame graphs, I expect the exact same stacks to be shown, just with leaves and roots swapped. So, when import system frames are not hidden, I expect us to show this:

| root size=2048 | | malloc size=2048 | | parent_one size=1024 | importer size=1024 | | main size=1024 | parent_two size=1024 | | | main size=1024 |

But when import system frames are hidden, I expect it to instead show:

| root size=2048 | | malloc size=1024 | parent_two size=1024 | | parent_one size=1024 | main size=1024 | | main size=1024 | |

The main -> parent_one -> malloc path remains unchanged by "hide import system frames", since that path doesn't go through the import system, but the main -> parent_two -> importer -> malloc stack gets truncated to main -> parent_two when importer is reached (just like it is for non-inverted flame graphs), and the stack we show now ends with parent_two as a child of the root (because parent_two is a leaf in the non-inverted hide-import-system case, and inverted mode swaps leaves and roots). Which means that the malloc node shrinks from 2048 bytes to 1024 bytes, since half of its allocations are attributable to a path through the import system, and half aren't.

There's no way we can achieve that desired behavior if we update the malloc node to say that it's an import system frame because one of its callers is an import system frame, even though the other isn't. As this function is currently implemented, I think it would result in us hiding malloc entirely, effectively truncating the main -> parent_one -> malloc call stack even though it didn't pass through the import system.

I can think of 3 reasonable ways to handle this:

Build two separate trees in --inverted mode. For one, use full stacks as the input. For the other, truncate the stacks at import system calls before processing them into a tree. Include both trees in the inverted-mode HTML, and have the checkbox switch which one is rendered.

In the FrameNodeDict, instead of recording just a bool for whether or not the node was reached by the import system, record an int: the number of bytes allocated at that node through a path that went through the import system. That would change this |= into a +=. When the checkbox is unset the inverted-mode JS code would ignore that field. When the checkbox is checked the JS code would shrink every node by treating that node's size as value - import_system_bytes, instead of as value bytes. Any node where value - import_system_bytes is 0 would be removed, and its children would be hoisted up to the root.

Pre-process each stack to identify which frames are below a call into the import system. Include the is-import-system flag as part of the node key, so that we have two different malloc nodes in the dumped tree, each associated with 1024 bytes. When the checkbox is checked, our JS code drops nodes with is-import-system set, and hoists their children up a level. When the checkbox is unchecked, our JS code would need to traverse the tree and merge nodes together if they have the same parent and same location, but one is flagged is-import-system and the other isn't.

Of these options, I think 1 is probably the simplest, and 3 is probably the most complex. 2 might require the fewest changes to what's been implemented so far...

@pablogsal Please check what I've said here, and see if a) you agree with me about what the checkbox should do, and b) if my proposed solutions seem reasonable.

godlygeek · 2023-08-21T22:40:39Z

src/memray/reporters/templates/__init__.py

+    if inverted and kind == "flamegraph":
+        kind = f"inverted {kind}"


Don't we want to return "inverted flamegraph report (memory leaks)" if show_memory_leaks is true? I think what we want is:

parts = [] if inverted: parts.append("inverted") parts.append(kind) parts.append("report") if show_memory_leaks: parts.append("(memory leaks)") return " ".join(parts)

Also: Don't worry about doing the and kind == "flamegraph" here. I think it's fine to return "inverted table report" if someone calls this with kind="table". We know they won't, but if they for some reason do, that would be sane behavior, so there's no reason to avoid it.

godlygeek · 2023-08-21T22:44:54Z

As a heads up, I haven't looked at the JS or the tests yet, so the above comments are what I have to say about the Python side of the implementation.

godlygeek

This is looking great! I've taken a look at the Javascript this time around (though I still haven't looked closely at the tests), and most of the concerns I have now are pretty minor.

godlygeek · 2023-09-01T00:14:13Z

src/memray/reporters/flamegraph.py

@@ -20,16 +29,43 @@
 from memray.reporters.frame_tools import is_cpython_internal
 from memray.reporters.frame_tools import is_frame_from_import_system
 from memray.reporters.frame_tools import is_frame_interesting
+from memray.reporters.stats import PythonStackElement


I'd rather not have one reporter importing things from another if we can avoid it. I'd rather copy-paste this type declaration, instead:

Suggested change

from memray.reporters.stats import PythonStackElement

PythonStackElement = Tuple[str, str, int]

That helps maintain the separation of responsibilities between different reporters. We might eventually want to move that definition into a common place that different reporters can share - maybe create src/memray/_types.py or something like that - but I wouldn't bother until we need it in yet another place.

godlygeek · 2023-09-01T00:18:46Z

src/memray/reporters/flamegraph.py

+    @classmethod
+    def update_import_system_frames_for_inverted_flamegraph(


This function is no longer used and should be removed

godlygeek · 2023-09-01T00:38:42Z

src/memray/reporters/flamegraph.py

+        if temporal and inverted:
+            raise NotImplementedError(
+                "FlamegraphReporter does not support temporal inverted flame graphs."
+            )


We should handle this in src/memray/commands/common.py in run instead, and do a parser.error() if the user provides both --temporal and --inverted. That'll give a nicer error message for users, I believe.

godlygeek · 2023-09-01T00:46:44Z

src/memray/reporters/flamegraph.py

+        frames: List[FrameNodeDict],
+        node_index_by_key: Dict[Tuple[int, StackFrame, str], int],
+        record: Union[AllocationRecord, TemporalAllocationRecord],
+        current_frame_id: int,


Callers always provide 0 for this argument. Since it doesn't vary from one call to the next, current_frame_id should be a local variable instead of a parameter.

godlygeek · 2023-09-01T00:49:51Z

src/memray/reporters/flamegraph.py

+                current_frame["location"] = ("...", "...", 0)
+                break
+
+        return frames


We don't need to return frames here. The caller passed frames to us, so they already have a reference to it. Returning it is misleading, since it isn't clear from the signature that we always return the same list, never a different one.

godlygeek · 2023-09-01T01:40:50Z

src/memray/reporters/flamegraph.py

+                import_system_stack = list(
+                    reversed(
+                        list(
+                            itertools.takewhile(
+                                lambda e: not is_frame_from_import_system(e),
+                                reversed(stack),
+                            )
                        )
                    )
+                )


I like this! Very elegant.

godlygeek · 2023-09-01T01:43:22Z

src/memray/reporters/flamegraph.py

            "value": 0,
            "children": [],
            "n_allocations": 0,
            "thread_id": "0x0",
            "interesting": True,
            "import_system": False,
        }
+        inverted_import_system_root = deepcopy(root) if inverted else None


I think every variable name in this function with _import_system_ in the name should be renamed to _no_import_system_ (or just _no_imports_) instead.

inverted_import_system_root makes it sound like this is the root of an inverted tree that does contain import system nodes, when actually its the root of an inverted tree that contains only non-import-system nodes.

godlygeek · 2023-09-01T01:49:15Z

src/memray/reporters/flamegraph.py

+        if temporal and inverted:
+            raise NotImplementedError(


Have you given any thought to what it would take to extend this to work for temporal mode? It should definitely be a separate PR, this one is long enough - but I don't think it would be too tough to implement.

godlygeek · 2023-09-01T01:54:02Z

src/memray/reporters/templates/base.html

@@ -139,8 +139,11 @@ <h5 class="modal-title" id="helpModalLabel">How to interpret {{ kind }} reports<
  <script type="text/javascript">
    const packed_data = {{ data|tojson }};
    var data = null;
+    var flamegraphData = null;
+    var invertedImportSystemData = null;


As above, this makes it sound like this one does include import system nodes, rather than does not. I'd call it something like importSystemHiddenFlamegraphData or noImportSystemData or flamegraphDataNoImportSystem or something.

godlygeek · 2023-09-01T01:57:59Z

src/memray/reporters/assets/flamegraph.js

+        nodes: inverted_import_system_nodes,
+        unique_threads: unique_threads,
+      })
+    : {};


Let's do null here instead of an empty object. That's a better indicator for the particular sort of invalid that this is.

Suggested change

: {};

: null;

godlygeek

I've made a few small changes, but this LGTM!

godlygeek · 2023-09-05T20:56:33Z

src/memray/reporters/flamegraph.py

+    size: Union[int, None]
+    n_allocations: Union[int, None]
+    intervals: Union[List[Interval], None]


Union[X, None] is the same thing as Optional[X], and we're already using Optional elsewhere in this file, so I'm going to update these 3 to match.

godlygeek · 2023-09-05T21:43:29Z

src/memray/reporters/flamegraph.py

+        cls,
+        frames: List[FrameNodeDict],
+        all_strings: StringRegistry,
+        interval_list: List[Tuple[int, Union[int, None], int, int, int]],


Since this function doesn't actually touch interval_list beyond seeing whether it exists, and since it exists if and only if we're in temporal mode, I'm going to switch this to just accepting temporal: bool instead.

godlygeek · 2023-09-05T21:56:07Z

src/memray/reporters/flamegraph.py

+        inverted_no_imports_frames: List[FrameNodeDict] = []
+
+        if inverted:
+            inverted_no_imports_root = deepcopy(root)


I don't like having root and inverted_no_imports_root as local variables in this function, since they're only used here at the very top of the fairly big function. Limiting the amount of local state will make this function easier to understand. I'm going to factor creation of the root node out into a function, and call it once for each tree we're making.

godlygeek · 2023-09-05T22:00:04Z

src/memray/reporters/flamegraph.py

        node_index_by_key: Dict[Tuple[int, StackFrame, str], int] = {}
+        inverted_no_imports_node_index_by_key: Dict[
+            Tuple[int, StackFrame, str], int
+        ] = {}


I'm gonna switch this to use a type alias, to make things slightly easier to read:

NodeKey = Tuple[int, StackFrame, str] node_index_by_key: Dict[NodeKey, int] = {} inverted_no_imports_node_index_by_key: Dict[NodeKey, int] = {}

In particular, it's tough to see that the two variables have the same type given the way that Black wrapped the lines, and this change makes that more recognizable.

godlygeek · 2023-09-05T22:03:06Z

src/memray/reporters/flamegraph.py

+            record_data: RecordData = {
+                "thread_name": thread_id,
+                "size": size,
+                "n_allocations": n_allocations,
+                "intervals": intervals,
+            }


I'm gonna move this to be defined in each branch of the if above, so that we don't need intervals and size and n_allocations and thread_id as local variables in addition to values in this dict, for the same reason as above - limiting the number of local variables here will help future readers.

godlygeek · 2023-09-05T22:09:06Z

src/memray/reporters/flamegraph.py

+                stack_it = enumerate(reversed(stack))
+                cls.generate_frames(
+                    stack_it=stack_it,


I'm gonna drop this local variable as well, and just construct it in the call to the function below, since it's only used in one spot, and to keep limiting local state.

godlygeek · 2023-09-05T22:10:07Z

src/memray/reporters/flamegraph.py

+                no_imports_stack = list(
+                    reversed(
+                        list(
+                            itertools.takewhile(
+                                lambda e: not is_frame_from_import_system(e),
+                                reversed(stack),
+                            )
                        )
                    )
+                )


I'm gonna split this into a function, so that we can assign a nice name to it for the benefit of future maintainers.

Added inverted flame graph which is generated with --inverted command line argument. Signed-off-by: Ivona Stojanovic <istojanovic2@bloomberg.net>

Added tests for inverted flame graph. Signed-off-by: Ivona Stojanovic <istojanovic2@bloomberg.net>

pablogsal requested review from pablogsal and godlygeek August 14, 2023 10:13

pablogsal reviewed Aug 14, 2023

View reviewed changes

sarahmonod mentioned this pull request Aug 15, 2023

Inverted flame graphs #293

Closed

1 task

ivonastojanovic force-pushed the inverted_flame_graph branch 2 times, most recently from 553d821 to 2abf102 Compare August 21, 2023 13:14

ivonastojanovic force-pushed the inverted_flame_graph branch from 2abf102 to a65cd31 Compare August 21, 2023 15:03

godlygeek reviewed Aug 21, 2023

View reviewed changes

godlygeek requested changes Aug 21, 2023

View reviewed changes

ivonastojanovic force-pushed the inverted_flame_graph branch from a65cd31 to 1c399a6 Compare August 30, 2023 14:19

godlygeek reviewed Sep 1, 2023

View reviewed changes

ivonastojanovic force-pushed the inverted_flame_graph branch from 1c399a6 to fd0b4f9 Compare September 1, 2023 16:44

godlygeek force-pushed the inverted_flame_graph branch 2 times, most recently from 91bc90e to 935184b Compare September 5, 2023 22:45

godlygeek approved these changes Sep 5, 2023

View reviewed changes

godlygeek force-pushed the inverted_flame_graph branch from 935184b to f99c6a8 Compare September 5, 2023 22:49

Ivona Stojanovic added 2 commits September 5, 2023 19:12

Inverted flame graph

9e314d8

Added inverted flame graph which is generated with --inverted command line argument. Signed-off-by: Ivona Stojanovic <istojanovic2@bloomberg.net>

tests: Inverted flame graph

e16d9b2

Added tests for inverted flame graph. Signed-off-by: Ivona Stojanovic <istojanovic2@bloomberg.net>

godlygeek force-pushed the inverted_flame_graph branch from f99c6a8 to e16d9b2 Compare September 5, 2023 23:12

godlygeek enabled auto-merge (rebase) September 5, 2023 23:12

godlygeek merged commit 68a453a into bloomberg:main Sep 5, 2023
33 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Inverted flame graph #439

Inverted flame graph #439

ivonastojanovic commented Aug 14, 2023 •

edited by sarahmonod

Loading

pablogsal Aug 14, 2023

pablogsal Aug 14, 2023

pablogsal Aug 14, 2023

pablogsal Aug 14, 2023

pablogsal Aug 14, 2023

codecov-commenter commented Aug 21, 2023 •

edited

Loading

godlygeek Aug 21, 2023 •

edited

Loading

godlygeek Aug 21, 2023

godlygeek Aug 21, 2023

godlygeek Aug 21, 2023

godlygeek Aug 21, 2023

godlygeek Aug 21, 2023

godlygeek Aug 21, 2023

godlygeek Aug 21, 2023

godlygeek commented Aug 21, 2023

godlygeek left a comment

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek Sep 1, 2023

godlygeek left a comment

godlygeek Sep 5, 2023

godlygeek Sep 5, 2023

godlygeek Sep 5, 2023

godlygeek Sep 5, 2023

godlygeek Sep 5, 2023

godlygeek Sep 5, 2023

godlygeek Sep 5, 2023

	if index - num_skipped_frames > MAX_STACKS:
	current_frame["name"] = "<STACK TOO DEEP>"
	current_frame["location"] = ["...", "...", 0]
	break

		@@ -0,0 +1,4 @@
		Add a new ``inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,

	Add a new ``inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,
	Add a new ``--inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,

		@@ -0,0 +1,4 @@
		Add a new ``inverted`` flag to ``memray flamegraph``. In an inverted flame graph, the roots are the functions that allocated memory,
		and the children of any given node represent the percentage of any of that node's

	allocations that can be attributed to a particular caller.The inverted flame graph is very helpful in analyzing where memory is being
	allocations that can be attributed to a particular caller. The inverted flame graph is very helpful in analyzing where memory is being

-        help="Invert flame graph",
+        help=(
+            "Invert the flame graph: "
+            "use allocators as roots instead of thread entry points",
+        )

	stack_it = enumerate((stack)) if inverted else enumerate(reversed(stack))
	stack_it = enumerate(stack) if inverted else enumerate(reversed(stack))

		if inverted and kind == "flamegraph":
		kind = f"inverted {kind}"

	from memray.reporters.stats import PythonStackElement
	PythonStackElement = Tuple[str, str, int]

		@classmethod
		def update_import_system_frames_for_inverted_flamegraph(

Inverted flame graph #439

Inverted flame graph #439

Conversation

ivonastojanovic commented Aug 14, 2023 • edited by sarahmonod Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov-commenter commented Aug 21, 2023 • edited Loading

Codecov Report

godlygeek Aug 21, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

godlygeek commented Aug 21, 2023

godlygeek left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

godlygeek left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ivonastojanovic commented Aug 14, 2023 •

edited by sarahmonod

Loading

codecov-commenter commented Aug 21, 2023 •

edited

Loading

godlygeek Aug 21, 2023 •

edited

Loading