Re: Parallel bitmap heap scan

On Mon, Feb 20, 2017 at 11:18 PM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Mon, Feb 20, 2017 at 5:55 AM, Thomas Munro
> <thomas(dot)munro(at)enterprisedb(dot)com> wrote:
>> On Sun, Feb 19, 2017 at 9:59 PM, Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:
>>> So basically, what I want to propose is that Only during
>>> ExecReScanBitmapHeapScan we can free all the DSA pointers because at
>>> that time we can be sure that all the workers have completed there
>>> task and we are safe to free. (And we don't free any DSA memory at
>>> ExecEndBitmapHeapScan).
>>
>> I think this works.
>
> OK.

In my latest version of the patch, I have fixed it as described above
i.e only cleanup in ExecReScanBitmapHeapScan.

For getting this there is some change in both the patches.

0001:
1. I have got a new interface, "tbm_free_shared_area(dsa_area *dsa,
dsa_pointer dp)" which will be responsible for freeing all the shared
members (pagetable, ptpage and ptchunk). Actually, we can not do this
in tbm_free itself because the only leader will have a local tbm with
reference to all these pointers and our parallel bitmap leader may not
necessarily be the actual backend.

If we want to get this done using tbm, then we need to create a local
tbm in each worker and get the shared member reference copied into tbm
using tbm_attach_shared_iterate like we were doing in the earlier
version.

2. Now tbm_free is not freeing any of the shared members which can be
accessed by other worker so tbm_free is safe to call from
ExecEndBitmapHeapScan without any safety check or ref count.

0002:
1. We don't need ExecShutdownBitmapHeapScan anymore because now we are
not freeing any shared member in ExecEndBitmapHeapScan.

2. In ExecReScanBitmapHeapScan we will call tbm_free_shared_area to
free the shared members of the TBM.
3. After that, we will free TBMSharedIteratorState what we allocated
using tbm_prepare_shared_iterate.

--
Regards,
Dilip Kumar
EnterpriseDB: http://www.enterprisedb.com