Tarantool development patches archive
 help / color / mirror / Atom feed
From: Sergey Kaplun via Tarantool-patches <tarantool-patches@dev.tarantool.org>
To: Sergey Ostanevich <sergos@tarantool.org>
Cc: tarantool-patches@dev.tarantool.org
Subject: Re: [Tarantool-patches] [PATCH luajit v1] tools: introduce --leak-only memprof parser option
Date: Wed, 17 Mar 2021 18:03:23 +0300	[thread overview]
Message-ID: <20210317150323.GB28016@root> (raw)
In-Reply-To: <D56710BE-9C25-4190-9026-EF240C007919@tarantool.org>

Hi!

Thanks for the review!

On 29.01.21, Sergey Ostanevich wrote:
> Hi! 
> Thanks for the patch, some comments below.
> 
> Regards,
> Sergos
> 
> 
> > On 25 Jan 2021, at 22:35, Sergey Kaplun <skaplun@tarantool.org> wrote:
> > 
> > This patch adds a new --leak-only memory profiler parser option.
> > When the parser runs with that option, it will report only lines
> > (or "INTERNAL") that allocate and do not free some amount of bytes.
> > The parser also reports the number of allocation and deallocation
> > events related to each line.
> > ---
> > 
> > Branch: https://github.com/tarantool/luajit/tree/skaplun/gh-noticket-memprof-memleaks-option
> > Testing branch: https://github.com/tarantool/tarantool/tree/skaplun/gh-noticket-memprof-memleaks-option

I removed the old branches and create issue[1]-related:
Branch: https://github.com/tarantool/luajit/tree/skaplun/gh-5812-memprof-memleaks-option
Testing branch: https://github.com/tarantool/tarantool/tree/skaplun/gh-5812-memprof-memleaks-option

> > 
> > Side note: I only update commit message thats why CI from the "old"
> > commit.
> > CI: https://gitlab.com/tarantool/tarantool/-/pipelines/246541599
> > 
> > This patch is a result of offline discussion of memory profiler
> > and its usage and feedback from Mons.
> > 
> > I don't know - is it reasonable to reference [1] issue here or create a
> > new one?
> 
> I think it’s ok to mention [1]

Updated commit message to the following:

| tools: introduce --leak-only memprof parser option
|
| Memprof parser now adds postamble with the source lines of Lua chunks
| (or "INTERNAL") that allocate and do not free some amount of bytes, when
| profiler finishes. The parser also reports the number of allocation and
| deallocation events related to each line.
|
| Also, this patch adds a new --leak-only memory profiler parser option.
| When the parser runs with that option, it reports only leak
| information.
|
| Resolves tarantool/tarantool#5812

Also, I've added the following patch to be able to run memprof from
Tarantool with leak-only option.

===================================================================
diff --git a/tools/memprof.lua b/tools/memprof.lua
index 85dadd4..d2231ee 100644
--- a/tools/memprof.lua
+++ b/tools/memprof.lua
@@ -107,10 +107,14 @@ local function dump(inputfile)
   os.exit(0)
 end
 
+local function dump_wrapped(...)
+  return dump(parseargs(...))
+end
+
 -- FIXME: this script should be application-independent.
 local args = {...}
 if #args == 1 and args[1] == "memprof" then
-  return dump
+  return dump_wrapped
 else
-  dump(parseargs(args))
+  dump_wrapped(args)
 end
===================================================================

Now memprof parser can be called from Tarantool with --leak-only option
like the following:

| $ src/tarantool -e 'local memprof = require("memprof")(arg)' - --leak-only /tmp/memprof.bin

Thoughts?

> 
> > 
> > The example of output. Assuming we have file <format_concat.lua>:
> > |  1 jit.off() -- more verbose reports
> > |  2
> > |  3 local function concat(a)
> > |  4   local nstr = a.."a"
> > |  5   return nstr
> > |  6 end
> > |  7
> > |  8 local function format(a)
> > |  9   local nstr = string.format("%sa", a)
> > | 10   return nstr
> > | 11 end
> > | 12
> > | 13 collectgarbage() -- cleanup
> > | 14
> > | 15 local binfile = "/tmp/memprof_"..(arg[0]):match("/([^/]*).lua")..".bin"
> > | 16
> > | 17 local st, err = misc.memprof.start(binfile)
> > | 18 assert(st, err)
> > | 19
> > | 20 for i = 1, 10000 do
> > | 21   local f = format(i)
> > | 22   local c = concat(i)
> > | 23 end
> > | 24
> > | 25 local st, err = misc.memprof.stop()
> > | 26 assert(st, err)
> > | 27
> > | 28 os.exit()
> > 
> > Parser's output without option:
> > | ALLOCATIONS
> > | @/home/burii/reports/demo_memprof/format_concat.lua:8, line 9: 19998	624322	0
> > | 
> > | REALLOCATIONS
> > | 
> > | DEALLOCATIONS
> > | INTERNAL: 283	0	5602
> > | 	Overrides:
> > | 		@/home/burii/reports/demo_memprof/format_concat.lua:8, line 9
> > | 
> > | @/home/burii/reports/demo_memprof/format_concat.lua:8, line 9: 2	0	98304
> > | 	Overrides:
> > | 		@/home/burii/reports/demo_memprof/format_concat.lua:8, line 9
> > 
> > And with:
> > 
> > | HEAP SUMMARY:
> > | @/home/burii/reports/demo_memprof/format_concat.lua:8, line 9 holds 553214 bytes 19998 allocs, 283 frees
> > 
> > Side note: The attentive reader will notice that we have 283+2 frees in
> > the first case and 283 in the second.
> > This is because --leak-only considers deallocation/reallocation events
> > for which we know the source address. The default report aggregates
> > these all deallocations into one. See `link_to_previous()` in
> > <tools/memprof/parse.lua> for details.
> > 
> > tools/memprof.lua          | 23 +++++-----
> > tools/memprof/humanize.lua | 91 +++++++++++++++++++++++++++++++++++++-
> > tools/memprof/parse.lua    | 20 +++++++--
> > 3 files changed, 115 insertions(+), 19 deletions(-)
> > 
> > diff --git a/tools/memprof.lua b/tools/memprof.lua
> > index 9f96208..5d09fbd 100644
> > --- a/tools/memprof.lua
> > +++ b/tools/memprof.lua
> > @@ -37,6 +37,11 @@ Supported options are:
> >   os.exit(0)
> > end
> > 
> > +local leak_only = false
> > +opt_map["leak-only"] = function()
> 
> Should you add this into the SYNOPSIS part?

Yes, thanks, see the iterative patch below:

===================================================================
diff --git a/tools/memprof.lua b/tools/memprof.lua
index 5d09fbd..e1af99f 100644
--- a/tools/memprof.lua
+++ b/tools/memprof.lua
@@ -33,6 +33,7 @@ luajit-parse-memprof [options] memprof.bin
 Supported options are:
 
   --help                            Show this help and exit
+  --leak-only                       Report only leaks information
 ]]
   os.exit(0)
 end
===================================================================

> 
> > +  leak_only = true
> > +end
> > +
> > -- Print error and exit with error status.
> > local function opterror(...)
> >   stderr:write("luajit-parse-memprof.lua: ERROR: ", ...)
> > @@ -94,19 +99,11 @@ local function dump(inputfile)
> >   local reader = bufread.new(inputfile)
> >   local symbols = symtab.parse(reader)
> >   local events = memprof.parse(reader, symbols)
> > -
> > -  stdout:write("ALLOCATIONS", "\n")
> > -  view.render(events.alloc, symbols)
> > -  stdout:write("\n")
> > -
> > -  stdout:write("REALLOCATIONS", "\n")
> > -  view.render(events.realloc, symbols)
> > -  stdout:write("\n")
> > -
> > -  stdout:write("DEALLOCATIONS", "\n")
> > -  view.render(events.free, symbols)
> > -  stdout:write("\n")
> > -
> 
> Should the HEAP SUMMARY appear in the end of the ‘all’ report?
> Otherwise the ‘_only’ in the option name is mesleading: show part of 
> the report, while it shows completely different one. Same for the
> code below.

Agree, fixed. Also, I renamed `all()` function name to the more verbose
one:

===================================================================
diff --git a/tools/memprof.lua b/tools/memprof.lua
index e1af99f..85dadd4 100644
--- a/tools/memprof.lua
+++ b/tools/memprof.lua
@@ -100,11 +100,10 @@ local function dump(inputfile)
   local reader = bufread.new(inputfile)
   local symbols = symtab.parse(reader)
   local events = memprof.parse(reader, symbols)
-  if leak_only then
-    view.leak_only(events, symbols)
-  else
-    view.all(events, symbols)
+  if not leak_only then
+    view.profile_info(events, symbols)
   end
+  view.leak_only(events, symbols)
   os.exit(0)
 end
 
diff --git a/tools/memprof/humanize.lua b/tools/memprof/humanize.lua
index bad0597..9e1bdf2 100644
--- a/tools/memprof/humanize.lua
+++ b/tools/memprof/humanize.lua
@@ -42,7 +42,7 @@ function M.render(events, symbols)
   end
 end
 
-function M.all(events, symbols)
+function M.profile_info(events, symbols)
   print("ALLOCATIONS")
   M.render(events.alloc, symbols)
   print("")
===================================================================

> 
> > +  if leak_only then
> > +    view.leak_only(events, symbols)
> > +  else
> > +    view.all(events, symbols)
> > +  end
> >   os.exit(0)
> > end
> > 
> > diff --git a/tools/memprof/humanize.lua b/tools/memprof/humanize.lua
> > index 109a39d..bad0597 100644
> > --- a/tools/memprof/humanize.lua
> > +++ b/tools/memprof/humanize.lua
> > @@ -28,8 +28,8 @@ function M.render(events, symbols)
> >     ))
> > 
> >     local prim_loc = {}
> > -    for _, loc in pairs(event.primary) do
> > -      table.insert(prim_loc, symtab.demangle(symbols, loc))
> > +    for _, heap_chunk in pairs(event.primary) do
> > +      table.insert(prim_loc, symtab.demangle(symbols, heap_chunk.loc))
> 
> The order of allocations is changed, was it incorrect in some way?

The `event.primary` array is modified and contains now not only location
line as a string, but a table with the line location in Lua chunk,
plus info about amount of allocated and freed bytes.

> 
> >     end
> >     if #prim_loc ~= 0 then
> >       table.sort(prim_loc)
> > @@ -42,4 +42,91 @@ function M.render(events, symbols)
> >   end
> > end
> > 
> > +function M.all(events, symbols)
> > +  print("ALLOCATIONS")
> > +  M.render(events.alloc, symbols)
> > +  print("")
> > +
> > +  print("REALLOCATIONS")
> > +  M.render(events.realloc, symbols)
> > +  print("")
> > +
> > +  print("DEALLOCATIONS")
> > +  M.render(events.free, symbols)
> > +  print("")
> > +end
> > +
> > +function M.leak_only(events, symbols)
> > +  -- Auto resurrects source event lines for counting/reporting.
> > +  local heap = setmetatable({}, {__index = function(t, line)
> > +    t[line] = {
> > +      size = 0,
> > +      cnt_alloc = 0,
> > +      cnt_free = 0,
> > +    }
> > +    return t[line]
> > +  end})
> > +
> > +  for _, event in pairs(events.alloc) do
> > +    if event.loc then
> > +      local ev_line = symtab.demangle(symbols, event.loc)
> > +
> > +      heap[ev_line].size = heap[ev_line].size + event.alloc
> > +      if (event.alloc > 0) then
> > +        heap[ev_line].cnt_alloc = heap[ev_line].cnt_alloc + event.num
> > +      end
> > +    end
> > +  end
> > +
> > +  -- Realloc and free events are pretty the same.
> > +  -- We don't interesting in aggregated alloc/free sizes for
>            aren’t interested

Fixed, thanks! See the iterative patch below.

> > +  -- the event, but only for new and old size values inside
> > +  -- alloc-realloc-free chain. Assuming that we have
> > +  -- no collisions between different object addresses.
> 
> I believe the address is source of the chunk {_|de|re}allocation, is there any
> different one?

I mean that, if some line allocated a few bytes in 0x4eadbeaf address,
freed and the other one line allocated data at the same address later,
it will be aggregated together.

> 
> > +  local function process_non_alloc_events(events_by_type)
> > +    for _, event in pairs(events_by_type) do
> > +      -- Realloc and free events always have "primary" key
> > +      -- that references table with rewrited memory
> > +      -- (may be empty).
> > +      for _, heap_chunk in pairs(event.primary) do
> > +        local ev_line = symtab.demangle(symbols, heap_chunk.loc)
> > +
> > +        heap[ev_line].size = heap[ev_line].size + heap_chunk.alloced
> > +        if (heap_chunk.alloced > 0) then
> 
> Can it be negative? If not - the above line can be put under this if also.

No, it can't, see the iterative patch below:

===================================================================
diff --git a/tools/memprof/humanize.lua b/tools/memprof/humanize.lua
index 9e1bdf2..b8dbfbf 100644
--- a/tools/memprof/humanize.lua
+++ b/tools/memprof/humanize.lua
@@ -71,15 +71,15 @@ function M.leak_only(events, symbols)
     if event.loc then
       local ev_line = symtab.demangle(symbols, event.loc)
 
-      heap[ev_line].size = heap[ev_line].size + event.alloc
       if (event.alloc > 0) then
+        heap[ev_line].size = heap[ev_line].size + event.alloc
         heap[ev_line].cnt_alloc = heap[ev_line].cnt_alloc + event.num
       end
     end
   end
 
   -- Realloc and free events are pretty the same.
-  -- We don't interesting in aggregated alloc/free sizes for
+  -- We aren't interested in aggregated alloc/free sizes for
   -- the event, but only for new and old size values inside
   -- alloc-realloc-free chain. Assuming that we have
   -- no collisions between different object addresses.
@@ -91,13 +91,13 @@ function M.leak_only(events, symbols)
       for _, heap_chunk in pairs(event.primary) do
         local ev_line = symtab.demangle(symbols, heap_chunk.loc)
 
-        heap[ev_line].size = heap[ev_line].size + heap_chunk.alloced
         if (heap_chunk.alloced > 0) then
+          heap[ev_line].size = heap[ev_line].size + heap_chunk.alloced
           heap[ev_line].cnt_alloc = heap[ev_line].cnt_alloc + heap_chunk.cnt
         end
 
-        heap[ev_line].size = heap[ev_line].size - heap_chunk.freed
         if (heap_chunk.freed > 0) then
+          heap[ev_line].size = heap[ev_line].size - heap_chunk.freed
           heap[ev_line].cnt_free = heap[ev_line].cnt_free + heap_chunk.cnt
         end
       end
===================================================================

> 
> > +          heap[ev_line].cnt_alloc = heap[ev_line].cnt_alloc + heap_chunk.cnt
> > +        end
> > +
> > +        heap[ev_line].size = heap[ev_line].size - heap_chunk.freed
> > +        if (heap_chunk.freed > 0) then
> > +          heap[ev_line].cnt_free = heap[ev_line].cnt_free + heap_chunk.cnt
> > +        end
> > +      end
> > +    end
> > +  end
> > +  process_non_alloc_events(events.realloc)
> > +  process_non_alloc_events(events.free)
> > +
> > +  local rest_heap = {}
> > +  for line, info in pairs(heap) do
> > +    -- Report "INTERNAL" events inconsistencies for profiling
> > +    -- with enabled jit.
> > +    if info.size > 0 then
> > +      table.insert(rest_heap, {line = line, hold_bytes = info.size})
> > +    end
> > +  end
> > +
> > +  table.sort(rest_heap, function(h1, h2)
> > +    return h1.hold_bytes > h2.hold_bytes
> > +  end)
> > +
> > +  print("HEAP SUMMARY:")
> > +  for _, h in pairs(rest_heap) do
> > +    print(string.format(
> > +      "%s holds %d bytes %d allocs, %d frees",
> > +      h.line, h.hold_bytes, heap[h.line].cnt_alloc, heap[h.line].cnt_free
> > +    ))
> > +  end
> > +  print("")
> > +end
> > +
> > return M
> > diff --git a/tools/memprof/parse.lua b/tools/memprof/parse.lua
> > index 6dae22d..374686c 100644
> > --- a/tools/memprof/parse.lua
> > +++ b/tools/memprof/parse.lua
> > @@ -39,11 +39,23 @@ local function new_event(loc)
> >   }
> > end
> > 
> > -local function link_to_previous(heap_chunk, e)
> > +local function link_to_previous(heap_chunk, e, nsize)
> >   -- Memory at this chunk was allocated before we start tracking.
> >   if heap_chunk then
> >     -- Save Lua code location (line) by address (id).
> > -    e.primary[heap_chunk[2]] = heap_chunk[3]
> > +    if e.primary[heap_chunk[2]] == nil then

I don't like this line now, so I've updated it like the following:

===================================================================
diff --git a/tools/memprof/parse.lua b/tools/memprof/parse.lua
index 374686c..df10a45 100644
--- a/tools/memprof/parse.lua
+++ b/tools/memprof/parse.lua
@@ -43,7 +43,7 @@ local function link_to_previous(heap_chunk, e, nsize)
   -- Memory at this chunk was allocated before we start tracking.
   if heap_chunk then
     -- Save Lua code location (line) by address (id).
-    if e.primary[heap_chunk[2]] == nil then
+    if not e.primary[heap_chunk[2]] then
       e.primary[heap_chunk[2]] = {
         loc = heap_chunk[3],
         alloced = 0,
===================================================================

> > +      e.primary[heap_chunk[2]] = {
> > +        loc = heap_chunk[3],
> > +        alloced = 0,
> > +        freed = 0,
> > +        cnt = 0,
> > +      }
> > +    end
> > +    -- Save memory diff heap information.
> > +    local location_data = e.primary[heap_chunk[2]]
> > +    location_data.alloced = location_data.alloced + nsize
> > +    location_data.freed = location_data.freed + heap_chunk[1]
> > +    location_data.cnt = location_data.cnt + 1
> >   end
> > end
> > 
> > @@ -97,7 +109,7 @@ local function parse_realloc(reader, asource, events, heap)
> >   e.free = e.free + osize
> >   e.alloc = e.alloc + nsize
> > 
> > -  link_to_previous(heap[oaddr], e)
> > +  link_to_previous(heap[oaddr], e, nsize)
> > 
> >   heap[oaddr] = nil
> >   heap[naddr] = {nsize, id, loc}
> > @@ -116,7 +128,7 @@ local function parse_free(reader, asource, events, heap)
> >   e.num = e.num + 1
> >   e.free = e.free + osize
> > 
> > -  link_to_previous(heap[oaddr], e)
> > +  link_to_previous(heap[oaddr], e, 0)
> > 
> >   heap[oaddr] = nil
> > end
> > -- 
> > 2.28.0
> > 
> > [1]: https://github.com/tarantool/tarantool/issues/5657
> 

[1]: https://github.com/tarantool/tarantool/issues/5812

-- 
Best regards,
Sergey Kaplun

  reply	other threads:[~2021-03-17 15:04 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-01-25 19:35 Sergey Kaplun via Tarantool-patches
2021-01-29 15:17 ` Sergey Ostanevich via Tarantool-patches
2021-03-17 15:03   ` Sergey Kaplun via Tarantool-patches [this message]
2021-03-29 15:50     ` Sergey Ostanevich via Tarantool-patches
2021-03-29  9:41 ` Igor Munkin via Tarantool-patches

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20210317150323.GB28016@root \
    --to=tarantool-patches@dev.tarantool.org \
    --cc=sergos@tarantool.org \
    --cc=skaplun@tarantool.org \
    --subject='Re: [Tarantool-patches] [PATCH luajit v1] tools: introduce --leak-only memprof parser option' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox