Content-Type: multipart/mixed; boundary="------------QjEeMMJzFkB0Si1E9GJBeB9f"
Message-ID: <f50cadde-6cf2-4dee-a2a9-75f274d4c0a0@yeah.net>
Date: Tue, 30 Jan 2024 21:25:35 +0800
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: Improvement discussion of custom and generic plans
To: pgsql-hackers@lists.postgresql.org
References: <9365b39d-a748-4e61-9396-466689ea0aa5@yeah.net>
Content-Language: en-US
From: Quan Zongliang <quanzongliang@yeah.net>
In-Reply-To: <9365b39d-a748-4e61-9396-466689ea0aa5@yeah.net>
Archived-At: 
 <https://www.postgresql.org/message-id/f50cadde-6cf2-4dee-a2a9-75f274d4c0a0%40yeah.net>
Precedence: bulk

This is a multi-part message in MIME format.
--------------QjEeMMJzFkB0Si1E9GJBeB9f
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit


On 2023/11/3 15:27, Quan Zongliang wrote:
> Hi
> 
> We have one such problem. A table field has skewed data. Statistics:
> n_distinct             | -0.4481973
> most_common_vals       | {5f006ca25b52ed78e457b150ee95a30c}
> most_common_freqs      | {0.5518474}
> 
> 
> Data generation:
> 
> CREATE TABLE s_user (
>       user_id varchar(32) NOT NULL,
>       corp_id varchar(32),
>       status int NOT NULL
>   );
> 
> insert into s_user
> select md5('user_id ' || a), md5('corp_id ' || a),
>       case random()<0.877675 when true then 1 else -1 end
>    FROM generate_series(1,10031) a;
> 
> insert into s_user
> select md5('user_id ' || a), md5('corp_id 10032'),
>       case random()<0.877675 when true then 1 else -1 end
>    FROM generate_series(10031,22383) a;
> 
> CREATE INDEX s_user_corp_id_idx ON s_user USING btree (corp_id);
> 
> analyze s_user;
> 
> 
> 1. First, define a PREPARE statement
> prepare stmt as select count(*) from s_user where status=1 and corp_id = 
> $1;
> 
> 2. Run it five times. Choose the custom plan.
> explain (analyze,buffers) execute stmt('5f006ca25b52ed78e457b150ee95a30c');
> 
> Here's the plan:
>   Aggregate  (cost=639.84..639.85 rows=1 width=8) (actual 
> time=4.653..4.654 rows=1 loops=1)
>     Buffers: shared hit=277
>     ->  Seq Scan on s_user  (cost=0.00..612.76 rows=10830 width=0) 
> (actual time=1.402..3.747 rows=10836 loops=1)
>           Filter: ((status = 1) AND ((corp_id)::text = 
> '5f006ca25b52ed78e457b150ee95a30c'::text))
>           Rows Removed by Filter: 11548
>           Buffers: shared hit=277
>   Planning Time: 0.100 ms
>   Execution Time: 4.674 ms
> (8 rows)
> 
> 3.From the sixth time. Choose generic plan.
> We can see that there is a huge deviation between the estimate and the 
> actual value:
>   Aggregate  (cost=11.83..11.84 rows=1 width=8) (actual 
> time=4.424..4.425 rows=1 loops=1)
>     Buffers: shared hit=154 read=13
>     ->  Bitmap Heap Scan on s_user  (cost=4.30..11.82 rows=2 width=0) 
> (actual time=0.664..3.371 rows=10836 loops=1)
>           Recheck Cond: ((corp_id)::text = $1)
>           Filter: (status = 1)
>           Rows Removed by Filter: 1517
>           Heap Blocks: exact=154
>           Buffers: shared hit=154 read=13
>           ->  Bitmap Index Scan on s_user_corp_id_idx  (cost=0.00..4.30 
> rows=2 width=0) (actual time=0.635..0.635 rows=12353 loops=1)
>                 Index Cond: ((corp_id)::text = $1)
>                 Buffers: shared read=13
>   Planning Time: 0.246 ms
>   Execution Time: 4.490 ms
> (13 rows)
> 
> This is because in the choose_custom_plan function, the generic plan is 
> attempted after executing the custom plan five times.
> 
>      if (plansource->num_custom_plans < 5)
>          return true;
> 
> The generic plan uses var_eq_non_const to estimate the average selectivity.
> 
> These are facts that many people already know. So a brief introduction.
> 
> 
> Our users actually use such parameter conditions in very complex PREPARE 
> statements. Once they use the generic plan for the sixth time. The 
> execution time will change from 5 milliseconds to 5 minutes.
> 
> 
> To improve this problem. The following approaches can be considered:
> 
> 1. Determine whether data skew exists in the PREPARE statement parameter 
> conditions based on the statistics.
> However, there is no way to know if the user will use the skewed parameter.
> 
> 2.When comparing the cost of the generic plan with the average cost of 
> the custom plan(function choose_custom_plan). Consider whether the 
> maximum cost of a custom plan executed is an order of magnitude 
> different from the cost of a generic plan.
> If the first five use a small selectivity condition. And after the sixth 
> use a high selectivity condition. Problems will still arise.
> 
> 3.Trace the execution time of the PREPARE statement. When an execution 
> time is found to be much longer than the average execution time, the 
> custom plan is forced to run.
> 
> 
> Is there any better idea?
> 
I tried to do a demo. Add a member paramid to Const. When Const is 
generated by Param, the Const is identified as coming from Param. Then 
check in var_eq_const to see if the field in the condition using this 
parameter is skewed. If so, choose_custom_plan returns true every time, 
forcing custom_plan to be used.
Only conditional expressions such as var eq param or param eq var can be 
supported.
If it makes sense. Continue to improve this patch.

> -- 
> Quan Zongliang
> 
> 
--------------QjEeMMJzFkB0Si1E9GJBeB9f
Content-Type: text/plain; charset=UTF-8; name="skewedparam.patch"
Content-Disposition: attachment; filename="skewedparam.patch"
Content-Transfer-Encoding: base64

ZGlmZiAtLWdpdCBhL3NyYy9iYWNrZW5kL29wdGltaXplci91dGlsL2NsYXVzZXMuYyBiL3Ny
Yy9iYWNrZW5kL29wdGltaXplci91dGlsL2NsYXVzZXMuYwppbmRleCA5NGViNTZhMWU3Li4z
Mzg0NTIwZGMxIDEwMDY0NAotLS0gYS9zcmMvYmFja2VuZC9vcHRpbWl6ZXIvdXRpbC9jbGF1
c2VzLmMKKysrIGIvc3JjL2JhY2tlbmQvb3B0aW1pemVyL3V0aWwvY2xhdXNlcy5jCkBAIC0y
NDg5LDYgKzI0ODksOCBAQCBldmFsX2NvbnN0X2V4cHJlc3Npb25zX211dGF0b3IoTm9kZSAq
bm9kZSwKIAkJCQkJCQkJCQkJcHZhbCwKIAkJCQkJCQkJCQkJcHJtLT5pc251bGwsCiAJCQkJ
CQkJCQkJCXR5cEJ5VmFsKTsKKwkJCQkJCQlpZiAocGFyYW1MSS0+cGFyYW1GZXRjaCA9PSBO
VUxMKQorCQkJCQkJCQljb24tPnBhcmFtaWQgPSBwYXJhbS0+cGFyYW1pZDsKIAkJCQkJCQlj
b24tPmxvY2F0aW9uID0gcGFyYW0tPmxvY2F0aW9uOwogCQkJCQkJCXJldHVybiAoTm9kZSAq
KSBjb247CiAJCQkJCQl9CmRpZmYgLS1naXQgYS9zcmMvYmFja2VuZC91dGlscy9hZHQvc2Vs
ZnVuY3MuYyBiL3NyYy9iYWNrZW5kL3V0aWxzL2FkdC9zZWxmdW5jcy5jCmluZGV4IGNlYTc3
N2U5ZDQuLjYyZGUxMjQ2NDMgMTAwNjQ0Ci0tLSBhL3NyYy9iYWNrZW5kL3V0aWxzL2FkdC9z
ZWxmdW5jcy5jCisrKyBiL3NyYy9iYWNrZW5kL3V0aWxzL2FkdC9zZWxmdW5jcy5jCkBAIC0y
NzMsMTAgKzI3MywzMiBAQCBlcXNlbF9pbnRlcm5hbChQR19GVU5DVElPTl9BUkdTLCBib29s
IG5lZ2F0ZSkKIAkgKiBpbiB0aGUgcXVlcnkuKQogCSAqLwogCWlmIChJc0Eob3RoZXIsIENv
bnN0KSkKLQkJc2VsZWMgPSB2YXJfZXFfY29uc3QoJnZhcmRhdGEsIG9wZXJhdG9yLCBjb2xs
YXRpb24sCi0JCQkJCQkJICgoQ29uc3QgKikgb3RoZXIpLT5jb25zdHZhbHVlLAotCQkJCQkJ
CSAoKENvbnN0ICopIG90aGVyKS0+Y29uc3Rpc251bGwsCi0JCQkJCQkJIHZhcm9ubGVmdCwg
bmVnYXRlKTsKKwl7CisJCUNvbnN0ICogY29uID0gKENvbnN0ICopIG90aGVyOworCisJCWlm
IChjb24tPnBhcmFtaWQgPiAwKQorCQl7CisJCQlib29sCXNrZXdlZF9zdGF0ID0gZmFsc2U7
CisKKwkJCXNlbGVjID0gdmFyX2VxX2NvbnN0X2V4dCgmdmFyZGF0YSwgb3BlcmF0b3IsIGNv
bGxhdGlvbiwKKwkJCQkJCQkJKChDb25zdCAqKSBvdGhlciktPmNvbnN0dmFsdWUsCisJCQkJ
CQkJCSgoQ29uc3QgKikgb3RoZXIpLT5jb25zdGlzbnVsbCwKKwkJCQkJCQkJdmFyb25sZWZ0
LCBuZWdhdGUsICZza2V3ZWRfc3RhdCk7CisKKwkJCWlmIChza2V3ZWRfc3RhdCkKKwkJCXsK
KwkJCQlQYXJhbUV4dGVybkRhdGEgKnBybTsKKworCQkJCXBybSA9ICZyb290LT5nbG9iLT5i
b3VuZFBhcmFtcy0+cGFyYW1zW2Nvbi0+cGFyYW1pZCAtIDFdOworCQkJCXBybS0+cGZsYWdz
IHw9IFBBUkFNX0ZMQUdfU0tFV0VEU1RBVDsKKwkJCX0KKwkJfQorCQllbHNlCisJCQlzZWxl
YyA9IHZhcl9lcV9jb25zdCgmdmFyZGF0YSwgb3BlcmF0b3IsIGNvbGxhdGlvbiwKKwkJCQkJ
CQkJKChDb25zdCAqKSBvdGhlciktPmNvbnN0dmFsdWUsCisJCQkJCQkJCSgoQ29uc3QgKikg
b3RoZXIpLT5jb25zdGlzbnVsbCwKKwkJCQkJCQkJdmFyb25sZWZ0LCBuZWdhdGUpOworCX0K
IAllbHNlCiAJCXNlbGVjID0gdmFyX2VxX25vbl9jb25zdCgmdmFyZGF0YSwgb3BlcmF0b3Is
IGNvbGxhdGlvbiwgb3RoZXIsCiAJCQkJCQkJCSB2YXJvbmxlZnQsIG5lZ2F0ZSk7CkBAIC0y
OTUsNiArMzE3LDE2IEBAIGRvdWJsZQogdmFyX2VxX2NvbnN0KFZhcmlhYmxlU3RhdERhdGEg
KnZhcmRhdGEsIE9pZCBvcHJvaWQsIE9pZCBjb2xsYXRpb24sCiAJCQkgRGF0dW0gY29uc3R2
YWwsIGJvb2wgY29uc3Rpc251bGwsCiAJCQkgYm9vbCB2YXJvbmxlZnQsIGJvb2wgbmVnYXRl
KQoreworCXJldHVybiB2YXJfZXFfY29uc3RfZXh0KHZhcmRhdGEsIG9wcm9pZCwgY29sbGF0
aW9uLAorCQkJIGNvbnN0dmFsLCBjb25zdGlzbnVsbCwKKwkJCSB2YXJvbmxlZnQsIG5lZ2F0
ZSwgTlVMTCk7Cit9CisKK2RvdWJsZQordmFyX2VxX2NvbnN0X2V4dChWYXJpYWJsZVN0YXRE
YXRhICp2YXJkYXRhLCBPaWQgb3Byb2lkLCBPaWQgY29sbGF0aW9uLAorCQkJIERhdHVtIGNv
bnN0dmFsLCBib29sIGNvbnN0aXNudWxsLAorCQkJIGJvb2wgdmFyb25sZWZ0LCBib29sIG5l
Z2F0ZSwgYm9vbCAqc2tld2VkX3N0YXQpCiB7CiAJZG91YmxlCQlzZWxlYzsKIAlkb3VibGUJ
CW51bGxmcmFjID0gMC4wOwpAQCAtMzg3LDYgKzQxOSwzNyBAQCB2YXJfZXFfY29uc3QoVmFy
aWFibGVTdGF0RGF0YSAqdmFyZGF0YSwgT2lkIG9wcm9pZCwgT2lkIGNvbGxhdGlvbiwKIAkJ
CQkJYnJlYWs7CiAJCQkJfQogCQkJfQorCisJCQlBc3NlcnQoc3Nsb3QubnZhbHVlcyA+IDAp
OworCQkJaWYgKHNrZXdlZF9zdGF0KQorCQkJeworCQkJCS8qIENoZWNrIHdoZXRoZXIgdGhl
IHZhbHVlcyBvZiB0aGUgZmllbGQgaXMgc2tld2VkIGFjY29yZGluZyB0byB0aGUgTUNWcy4g
Ki8KKwkJCQlpZiAoc3Nsb3QubnZhbHVlcyA+IDEgJiYKKwkJCQkJCXNzbG90Lm51bWJlcnNb
MF0gPiBzc2xvdC5udW1iZXJzW3NzbG90Lm52YWx1ZXMgLSAxXSAqIDEwKQorCQkJCQkvKiBD
b21wYXJlIHRoZSBtYXhpbXVtIGFuZCBtaW5pbXVtIHF1YW50aXRpZXMgaW4gdGhlIE1DVnMu
ICovCisJCQkJCSpza2V3ZWRfc3RhdCA9IHRydWU7CisJCQkJZWxzZQorCQkJCXsKKwkJCQkJ
LyoKKwkJCQkJICogQ29tcGFyZSB0aGUgbWF4aW11bSBudW1iZXIgaW4gdGhlIE1DVnMgd2l0
aCB0aGUgYXZlcmFnZSBudW1iZXIgb2YKKwkJCQkJICogb3RoZXIgdmFsdWVzKG5vdCBpbiB0
aGUgTUNWcykuCisJCQkJCSAqLworCQkJCQlkb3VibGUJCXN1bWNvbW1vbiA9IDAuMDsKKwkJ
CQkJZG91YmxlCQlvdGhlcmRpc3RpbmN0OworCisJCQkJCWZvciAoaSA9IDA7IGkgPCBzc2xv
dC5ubnVtYmVyczsgaSsrKQorCQkJCQkJc3VtY29tbW9uICs9IHNzbG90Lm51bWJlcnNbaV07
CisKKwkJCQkJb3RoZXJkaXN0aW5jdCA9IGdldF92YXJpYWJsZV9udW1kaXN0aW5jdCh2YXJk
YXRhLCAmaXNkZWZhdWx0KSAtCisJCQkJCQlzc2xvdC5ubnVtYmVyczsKKworCQkJCQlpZiAo
b3RoZXJkaXN0aW5jdCA+IDApCisJCQkJCXsKKwkJCQkJCWlmIChzc2xvdC5udW1iZXJzWzBd
ID4gKCgxLjAgLSBzdW1jb21tb24gLSBudWxsZnJhYykgLyBvdGhlcmRpc3RpbmN0KSAqIDEw
KQorCQkJCQkJCSpza2V3ZWRfc3RhdCA9IHRydWU7CisJCQkJCX0KKwkJCQl9CisJCQl9CiAJ
CX0KIAkJZWxzZQogCQl7CmRpZmYgLS1naXQgYS9zcmMvYmFja2VuZC91dGlscy9jYWNoZS9w
bGFuY2FjaGUuYyBiL3NyYy9iYWNrZW5kL3V0aWxzL2NhY2hlL3BsYW5jYWNoZS5jCmluZGV4
IDUxOTRjYmYyY2MuLjQwMTQxODMxOWIgMTAwNjQ0Ci0tLSBhL3NyYy9iYWNrZW5kL3V0aWxz
L2NhY2hlL3BsYW5jYWNoZS5jCisrKyBiL3NyYy9iYWNrZW5kL3V0aWxzL2NhY2hlL3BsYW5j
YWNoZS5jCkBAIC05OTEsNiArOTkxLDIwIEBAIEJ1aWxkQ2FjaGVkUGxhbihDYWNoZWRQbGFu
U291cmNlICpwbGFuc291cmNlLCBMaXN0ICpxbGlzdCwKIAllbHNlCiAJCXBsYW5fY29udGV4
dCA9IEN1cnJlbnRNZW1vcnlDb250ZXh0OwogCisJaWYgKGJvdW5kUGFyYW1zKQorCXsKKwkJ
aW50IGk7CisKKwkJZm9yIChpID0wOyBpIDwgYm91bmRQYXJhbXMtPm51bVBhcmFtczsgaSsr
KQorCQl7CisJCQlpZiAoYm91bmRQYXJhbXMtPnBhcmFtc1tpXS5wZmxhZ3MgJiBQQVJBTV9G
TEFHX1NLRVdFRFNUQVQpCisJCQl7CisJCQkJcGxhbnNvdXJjZS0+aGFzU2tld2VkcGFyYW0g
PSB0cnVlOworCQkJCWJyZWFrOworCQkJfQorCQl9CisJfQorCiAJLyoKIAkgKiBDcmVhdGUg
YW5kIGZpbGwgdGhlIENhY2hlZFBsYW4gc3RydWN0IHdpdGhpbiB0aGUgbmV3IGNvbnRleHQu
CiAJICovCkBAIC0xMDY2LDYgKzEwODAsOSBAQCBjaG9vc2VfY3VzdG9tX3BsYW4oQ2FjaGVk
UGxhblNvdXJjZSAqcGxhbnNvdXJjZSwgUGFyYW1MaXN0SW5mbyBib3VuZFBhcmFtcykKIAlp
ZiAocGxhbl9jYWNoZV9tb2RlID09IFBMQU5fQ0FDSEVfTU9ERV9GT1JDRV9DVVNUT01fUExB
TikKIAkJcmV0dXJuIHRydWU7CiAKKwlpZiAocGxhbnNvdXJjZS0+aGFzU2tld2VkcGFyYW0p
CisJCXJldHVybiB0cnVlOworCiAJLyogU2VlIGlmIGNhbGxlciB3YW50cyB0byBmb3JjZSB0
aGUgZGVjaXNpb24gKi8KIAlpZiAocGxhbnNvdXJjZS0+Y3Vyc29yX29wdGlvbnMgJiBDVVJT
T1JfT1BUX0dFTkVSSUNfUExBTikKIAkJcmV0dXJuIGZhbHNlOwpkaWZmIC0tZ2l0IGEvc3Jj
L2luY2x1ZGUvTWFrZWZpbGUgYi9zcmMvaW5jbHVkZS9NYWtlZmlsZQppbmRleCA1ODRkMDMw
MGQ5Li4wODBiZjgxODY3IDEwMDY0NAotLS0gYS9zcmMvaW5jbHVkZS9NYWtlZmlsZQorKysg
Yi9zcmMvaW5jbHVkZS9NYWtlZmlsZQpAQCAtNTEsNiArNTEsNyBAQCBpbnN0YWxsOiBhbGwg
aW5zdGFsbGRpcnMKIAkkKElOU1RBTExfREFUQSkgdXRpbHMvZm1ncnByb3Rvcy5oICckKERF
U1RESVIpJChpbmNsdWRlZGlyX3NlcnZlcikvdXRpbHMnCiAJJChJTlNUQUxMX0RBVEEpICQo
c3JjZGlyKS8qLmggJyQoREVTVERJUikkKGluY2x1ZGVkaXJfc2VydmVyKScKIAlmb3IgZGly
IGluICQoU1VCRElSUyk7IGRvIFwKKwkgIHB3ZCBcCiAJICAkKElOU1RBTExfREFUQSkgJChz
cmNkaXIpLyQkZGlyLyouaCAnJChERVNURElSKSQoaW5jbHVkZWRpcl9zZXJ2ZXIpJy8kJGRp
ciB8fCBleGl0OyBcCiAJZG9uZQogaWZlcSAoJCh2cGF0aF9idWlsZCkseWVzKQpkaWZmIC0t
Z2l0IGEvc3JjL2luY2x1ZGUvbm9kZXMvcGFyYW1zLmggYi9zcmMvaW5jbHVkZS9ub2Rlcy9w
YXJhbXMuaAppbmRleCAxMGM5ZmM1NDEzLi4wNTFhMTJiYjFkIDEwMDY0NAotLS0gYS9zcmMv
aW5jbHVkZS9ub2Rlcy9wYXJhbXMuaAorKysgYi9zcmMvaW5jbHVkZS9ub2Rlcy9wYXJhbXMu
aApAQCAtODUsNyArODUsMTMgQEAgc3RydWN0IFBhcnNlU3RhdGU7CiAgKgkgIGFuZCBwYXJh
bUNvbXBpbGVBcmcgaXMgcmF0aGVyIGFyYml0cmFyeS4KICAqLwogCi0jZGVmaW5lIFBBUkFN
X0ZMQUdfQ09OU1QJMHgwMDAxCS8qIHBhcmFtZXRlciBpcyBjb25zdGFudCAqLworLyogcGFy
YW1ldGVyIGlzIGNvbnN0YW50ICovCisjZGVmaW5lIFBBUkFNX0ZMQUdfQ09OU1QJCQkweDAw
MDEKKy8qCisgKiBUaGUgdmFyIHNpZGUgb2YgJ3ZhciBlcSBwYXJhbScgb3IgJ3BhcmFtIGVx
IHZhcicgZXhwcmVzc2lvbgorICogY29ycmVzcG9uZGluZyB0byB0aGUgcGFyYW1ldGVyIGNv
bnRhaW5zIHNrZXdlZCBkYXRhLgorICovCisjZGVmaW5lIFBBUkFNX0ZMQUdfU0tFV0VEU1RB
VAkJMHgwMDAyCiAKIHR5cGVkZWYgc3RydWN0IFBhcmFtRXh0ZXJuRGF0YQogewpkaWZmIC0t
Z2l0IGEvc3JjL2luY2x1ZGUvbm9kZXMvcHJpbW5vZGVzLmggYi9zcmMvaW5jbHVkZS9ub2Rl
cy9wcmltbm9kZXMuaAppbmRleCA0YTE1NDYwNmQyLi43ZTY3NjJkZTQ3IDEwMDY0NAotLS0g
YS9zcmMvaW5jbHVkZS9ub2Rlcy9wcmltbm9kZXMuaAorKysgYi9zcmMvaW5jbHVkZS9ub2Rl
cy9wcmltbm9kZXMuaApAQCAtMzE0LDYgKzMxNCwxMiBAQCB0eXBlZGVmIHN0cnVjdCBDb25z
dAogCSAqLwogCWJvb2wJCWNvbnN0Ynl2YWwgcGdfbm9kZV9hdHRyKHF1ZXJ5X2p1bWJsZV9p
Z25vcmUpOwogCisJLyoKKwkgKiA+MCBpbmRpY2F0ZXMgdGhhdCB0aGUgY29uc3RhbnQgaXMg
ZXh0cmFjdGVkIGZyb20gUGFyYW0uCisJICogSXRzIHZhbHVlIGlzIHBhcmFtaWQgb2YgdGhl
IFBhcmFtIHN0cnVjdC4KKwkgKi8KKwlpbnQJCQlwYXJhbWlkIHBnX25vZGVfYXR0cihxdWVy
eV9qdW1ibGVfaWdub3JlKTsKKwogCS8qCiAJICogdG9rZW4gbG9jYXRpb24sIG9yIC0xIGlm
IHVua25vd24uICBBbGwgY29uc3RhbnRzIGFyZSB0cmFja2VkIGFzCiAJICogbG9jYXRpb25z
IGluIHF1ZXJ5IGp1bWJsaW5nLCB0byBiZSBtYXJrZWQgYXMgcGFyYW1ldGVycy4KZGlmZiAt
LWdpdCBhL3NyYy9pbmNsdWRlL3V0aWxzL3BsYW5jYWNoZS5oIGIvc3JjL2luY2x1ZGUvdXRp
bHMvcGxhbmNhY2hlLmgKaW5kZXggYTkwZGZkZjkwNi4uNzEyMDllMDI5ZiAxMDA2NDQKLS0t
IGEvc3JjL2luY2x1ZGUvdXRpbHMvcGxhbmNhY2hlLmgKKysrIGIvc3JjL2luY2x1ZGUvdXRp
bHMvcGxhbmNhY2hlLmgKQEAgLTEyNCw2ICsxMjQsNyBAQCB0eXBlZGVmIHN0cnVjdCBDYWNo
ZWRQbGFuU291cmNlCiAJYm9vbAkJaXNfY29tcGxldGU7CS8qIGhhcyBDb21wbGV0ZUNhY2hl
ZFBsYW4gYmVlbiBkb25lPyAqLwogCWJvb2wJCWlzX3NhdmVkOwkJLyogaGFzIENhY2hlZFBs
YW5Tb3VyY2UgYmVlbiAic2F2ZWQiPyAqLwogCWJvb2wJCWlzX3ZhbGlkOwkJLyogaXMgdGhl
IHF1ZXJ5X2xpc3QgY3VycmVudGx5IHZhbGlkPyAqLworCWJvb2wJCWhhc1NrZXdlZHBhcmFt
OyAvKiBzZWUgdGhlIGRlZmluaXRpb24gb2YgUEFSQU0gRkxBRyBTS0VXRURTVEFUICovCiAJ
aW50CQkJZ2VuZXJhdGlvbjsJCS8qIGluY3JlbWVudHMgZWFjaCB0aW1lIHdlIGNyZWF0ZSBh
IHBsYW4gKi8KIAkvKiBJZiBDYWNoZWRQbGFuU291cmNlIGhhcyBiZWVuIHNhdmVkLCBpdCBp
cyBhIG1lbWJlciBvZiBhIGdsb2JhbCBsaXN0ICovCiAJZGxpc3Rfbm9kZQlub2RlOwkJCS8q
IGxpc3QgbGluaywgaWYgaXNfc2F2ZWQgKi8KZGlmZiAtLWdpdCBhL3NyYy9pbmNsdWRlL3V0
aWxzL3NlbGZ1bmNzLmggYi9zcmMvaW5jbHVkZS91dGlscy9zZWxmdW5jcy5oCmluZGV4IDJm
YTRjNGZjMWIuLmE3NTMyYjhkN2IgMTAwNjQ0Ci0tLSBhL3NyYy9pbmNsdWRlL3V0aWxzL3Nl
bGZ1bmNzLmgKKysrIGIvc3JjL2luY2x1ZGUvdXRpbHMvc2VsZnVuY3MuaApAQCAtMTg0LDYg
KzE4NCwxMCBAQCBleHRlcm4gZG91YmxlIHZhcl9lcV9jb25zdChWYXJpYWJsZVN0YXREYXRh
ICp2YXJkYXRhLAogCQkJCQkJICAgT2lkIG9wcm9pZCwgT2lkIGNvbGxhdGlvbiwKIAkJCQkJ
CSAgIERhdHVtIGNvbnN0dmFsLCBib29sIGNvbnN0aXNudWxsLAogCQkJCQkJICAgYm9vbCB2
YXJvbmxlZnQsIGJvb2wgbmVnYXRlKTsKK2V4dGVybiBkb3VibGUgdmFyX2VxX2NvbnN0X2V4
dChWYXJpYWJsZVN0YXREYXRhICp2YXJkYXRhLAorCQkJCQkJICAgT2lkIG9wcm9pZCwgT2lk
IGNvbGxhdGlvbiwKKwkJCQkJCSAgIERhdHVtIGNvbnN0dmFsLCBib29sIGNvbnN0aXNudWxs
LAorCQkJCQkJICAgYm9vbCB2YXJvbmxlZnQsIGJvb2wgbmVnYXRlLCBib29sICpza2V3ZWRf
c3RhdCk7CiBleHRlcm4gZG91YmxlIHZhcl9lcV9ub25fY29uc3QoVmFyaWFibGVTdGF0RGF0
YSAqdmFyZGF0YSwKIAkJCQkJCQkgICBPaWQgb3Byb2lkLCBPaWQgY29sbGF0aW9uLAogCQkJ
CQkJCSAgIE5vZGUgKm90aGVyLAo=

--------------QjEeMMJzFkB0Si1E9GJBeB9f--