Possible memory leak when use PooledDbContextFactory

[FarseerSky]

Issue Description:
I have encountered a memory dump issue during my stress test. The memory dump shows that there are 95000 instances of EntityFrameworkCore.Update.ColumnModification held by 239 DbContext objects in the pool.

I have investigated the source code and found that in the TopologicalSort method of CommandBatchPreparer, it clears the _modificationCommandGraph field which stores all the commands from the last time only when a new BatchCommands invoked. Can this be cleared after the SaveChangesAsync() method instead?

Code snippet:

await using var ctx = await _contextFactory.CreateDbContextAsync();
foreach (var bean in beans)
{
    ctx.Update(bean);
}

var res = await ctx.SaveChangesAsync();
ctx.ChangeTracker.Clear();

System Information:

• EF Core version:8.0.0
• Database provider: Pomelo.EntityFrameworkCore.MySql 8.0.0-beta.2
• Target framework: NET 8.0
• Operating system:Linux

[ajcvickers]

/cc @AndriySvyryd

[MoMack20]

Will this make it into a EF Core 8 patch? If not, is the call to ctx.ChangeTracker.Clear(); enough to mitigate this issue?

[mu88]

@MoMack20 no, unfortunately, calling context.ChangeTracker.Clear(); is not enough to mitigate the issue as you can see in my other issue

[FarseerSky]

@mu88 Looking at the milestone, it should be fixed in version 9.0. I have given up using PooledDbContext, and the non-pooled performance is also very good.