[1/2] read-cache: free threaded memory pool

Commit Message

Derrick Stolee Sept. 30, 2024, 11:40 a.m. UTC

From: Derrick Stolee <[e-mail]>

In load_cache_entries_threaded(), each thread is allocated its own
memory pool. This pool needs to be cleaned up while closing the threads
down, or it will be leaked.

Signed-off-by: Derrick Stolee <[e-mail]>
---
 read-cache.c | 1 +
 1 file changed, 1 insertion(+)

Comments

Patrick Steinhardt Sept. 30, 2024, 12:32 p.m. UTC | #1

On Mon, Sep 30, 2024 at 11:40:23AM +0000, Derrick Stolee via GitGitGadget wrote:
> From: Derrick Stolee <[e-mail]>
> 
> In load_cache_entries_threaded(), each thread is allocated its own

s/allocated/allocating/

> memory pool. This pool needs to be cleaned up while closing the threads
> down, or it will be leaked.
> 
> Signed-off-by: Derrick Stolee <[e-mail]>
> ---
>  read-cache.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/read-cache.c b/read-cache.c
> index 764fdfec465..3c078afadbc 100644
> --- a/read-cache.c
> +++ b/read-cache.c
> @@ -2188,6 +2188,7 @@ static unsigned long load_cache_entries_threaded(struct index_state *istate, con
>  		if (err)
>  			die(_("unable to join load_cache_entries thread: %s"), strerror(err));
>  		mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
> +		free(p->ce_mem_pool);
>  		consumed += p->consumed;
>  	}

Okay. We move over the contents of the pool, but forgot to free the pool
itself. As far as I can see the pool is always allocated and only used
in two functions, both of which assume that it is allocated. So I wonder
why it is allocated in the first place instead of making it a direct
member of `struct load_cache_entries_thread_data`.

Patrick

Derrick Stolee Oct. 1, 2024, 1:20 p.m. UTC | #2

On 9/30/24 8:32 AM, Patrick Steinhardt wrote:
> On Mon, Sep 30, 2024 at 11:40:23AM +0000, Derrick Stolee via GitGitGadget wrote:
>> From: Derrick Stolee <[e-mail]>
>>
>> In load_cache_entries_threaded(), each thread is allocated its own
> 
> s/allocated/allocating/

You're right that the wording is awkward but I'm not thrilled with the
suggested alternative.

Perhaps "each thread allocates its own"

>> memory pool. This pool needs to be cleaned up while closing the threads
>> down, or it will be leaked.

> Okay. We move over the contents of the pool, but forgot to free the pool
> itself. As far as I can see the pool is always allocated and only used
> in two functions, both of which assume that it is allocated. So I wonder
> why it is allocated in the first place instead of making it a direct
> member of `struct load_cache_entries_thread_data`.

I took a look at what it would take to replace the pointer with an inline
struct but found complications with situations such as the find_mem_pool()
method. While we could replace some of the logic to recognize the new
type, the existing logic seems to depend on using the NULL pointer as an
indicator that the pool should be lazily initialized.

If we were to pull the struct inline, we would either need another boolean
to indicate initialization or lose lazy initialization.

I'm leaning towards the simpler leak fix over the disruption of that
change.

Thanks,
-Stolee

Patrick Steinhardt Oct. 1, 2024, 2:11 p.m. UTC | #3

On Tue, Oct 01, 2024 at 09:20:01AM -0400, Derrick Stolee wrote:
> On 9/30/24 8:32 AM, Patrick Steinhardt wrote:
> > On Mon, Sep 30, 2024 at 11:40:23AM +0000, Derrick Stolee via GitGitGadget wrote:
> > > From: Derrick Stolee <[e-mail]>
> > > 
> > > In load_cache_entries_threaded(), each thread is allocated its own
> > 
> > s/allocated/allocating/
> 
> You're right that the wording is awkward but I'm not thrilled with the
> suggested alternative.
> 
> Perhaps "each thread allocates its own"

Sure, works for me :)

> > > memory pool. This pool needs to be cleaned up while closing the threads
> > > down, or it will be leaked.
> 
> > Okay. We move over the contents of the pool, but forgot to free the pool
> > itself. As far as I can see the pool is always allocated and only used
> > in two functions, both of which assume that it is allocated. So I wonder
> > why it is allocated in the first place instead of making it a direct
> > member of `struct load_cache_entries_thread_data`.
> 
> I took a look at what it would take to replace the pointer with an inline
> struct but found complications with situations such as the find_mem_pool()
> method. While we could replace some of the logic to recognize the new
> type, the existing logic seems to depend on using the NULL pointer as an
> indicator that the pool should be lazily initialized.
> 
> If we were to pull the struct inline, we would either need another boolean
> to indicate initialization or lose lazy initialization.
> 
> I'm leaning towards the simpler leak fix over the disruption of that
> change.

Fair enough, no complaint from my side. I thought it would've been easy,
but didn't dive deep. So if you say it is harder than I made it out to
be with my shallow understanding I'm going to trust your judgement.
After all, the leak fix is a strict improvement by itself.

Patrick

Patch

diff --git a/read-cache.c b/read-cache.c
index 764fdfec465..3c078afadbc 100644
--- a/read-cache.c
+++ b/read-cache.c
@@ -2188,6 +2188,7 @@  static unsigned long load_cache_entries_threaded(struct index_state *istate, con
 		if (err)
 			die(_("unable to join load_cache_entries thread: %s"), strerror(err));
 		mem_pool_combine(istate->ce_mem_pool, p->ce_mem_pool);
+		free(p->ce_mem_pool);
 		consumed += p->consumed;
 	}