Has my Plan changed or not?
We have learned from Kerry Osborne’s recent post on SQL Gone Bad – But Plan Not Changed?, and a prior post from Randolf Geist on PLAN_HASH_VALUE – How equal (and stable?) are your execution plans – part 1, that having the same Plan Hash Value (PHV) it is not guarantee that we actually have the same Execution Plan if you were to consider Access and Filter Predicates for Plan Operations. So, what does it mean? It means that if you have inconsistent performance out of the same Execution Plan (same PHV) it is not enough to just check if you have or not the same PHV. You need to be aware of the Predicates as well. Of course there are other possible explanations for having inconsistent performance out of the same PHV, like skewed data combined with the use of binds, but that is another story.
Same Plan but different PHV
The opposite to “Same PHV but different Plan” is also possible. So, after we understood we can have the same PHV but not quite the same Plan if we include in the compare the Predicates, the question became: Can we also have cases having the same Plan (including Predicates) but get a different PHV? And the answer is YES.
Discard the keyword STORAGE, which we can consider not strong enough to determine a Plan Operation is different. Of course this is questionable, but as of today the PHV is blind to the STORAGE keyword, so a Plan with or without this keyword would render the same PHV.
What about system-generated names like ‘SYS_TEMP%’, ‘index$_join$_%’ and ‘VW_ST%’? In all these 3 cases the PHV will be different but the Plan is actually the same, with the exception of the system-generated names. So, if you just look at the PHV and see that is different then it is also possible that actually you have the same Plan.
What if in the case of having the same index name, the set of columns or their order is different? In these cases you may look at the PHV and see the same value, and indeed it is the same Plan, but if the columns on a referenced index have changed, is it really the same Plan? In my opinion it is not! You could be spinning on an issue where you have same Plan, different Performance, but an Index changed its columns.
SQLT to the rescue
Since we have the two cases: “Same PHV but different Plan” and “Same Plan but different PHV”, reviewing and ruling out these two possible cases on a complex Execution Plan can be cumbersome and time consuming. That is WHY SQLT incorporated the concept of SQLT Plan Hash Value (SQLT_PHV) since version 126.96.36.199 (from May 20, 2010). First came SQLT_PHV1, then SQLT_PHV2 (on SQLT 188.8.131.52 on February 18, 2011). So we have now PHV, SQLT_PHV1 and SQLT_PHV2, as you can see below.
As you can see in Table foot note: SQLT_PHV1 considers id, parent_id, operation, options, index_columns and object_name. SQLT PHV2 includes also access and filter predicates. So when comparing Plans the values of PHV, SQLT_PHV1 and SQLT_PHV2 can give you a better idea if your Plan is actually the same or not. SQLT COMPARE also uses the 3 values to determine if your Plan is the same or not, and in cases like the PHV is the same but a Predicate on a Plan Operation is different, it highlights in red the specific Plan Operation that has a different Predicate. Pretty cool right? I guess I will have to blog about SQLT COMPARE some time soon then…
Oh, be also aware that AWR does not store Plan Predicates, so if your Plan only exists on AWR you may be blind to these Predicates, but if the EXPLAIN PLAN FOR renders the same Plan as seen in lines 4 and 5 above, then you can see the predicates for 657302870 out the “XPL” Plan instead of “AWR”. A little trick that becomes handy…
When it comes to Execution Plans and their Plan Hash Value, it is possible that two Plans with same PHV are actually different if you consider the Predicates, and also possible you get a different PHV even if the Plan is the same if your Plan has system-generated names. So, during your analysis just looking at the PHV to determine if two Plans are the same or not is not enough. If you are already using SQLT, pay attention to the “Execution Plans” section, and to the SQLT PHV1 and PHV2 columns.