What is persisted index?

[Former Member]

Note 1976994 - Wrong data results on persisted indexes mentioned a new term : persisted index.

I have searched OSS and Google, there is no thread about it except this note.

If it is an index persisted to the harddisk, what is the name convention for that object?

If it is persisted to the file system, what is its location?

Thanks a lot!

[Former Member · Accepted solution]

Honestly I have no idea. Perhaps knows. He knows this sort of gnarly thing. My read is that it was a feature that SAP didn't intend to release, but I don't know for sure.

[lbreddemann · Accepted solution]

Hmm.. hmm... weired things happening here in SCN.

The Rules of Engagement (http://scn.sap.com/docs/DOC-18590) document is not accessible anymore??

Anyhow, didn't we have that rule to use our real names here?

If that's your real name "cc 122010" I am actually sorry for you.

So, persisted indexes?

It's a rather straight forward thingy. Up until SPS 7 the default behavior for inverted indexes on column store table columns was: re-create them after loading the column to memory.

As all the information we need to create the index is in that column, there is no need to store it on disk.

That's fine until you have

a) a rather large indexes that take long to create (keep in mind, the inverted indexes are pretty similar to the main/value vector and compression is applied to the inverted index as well)

and/or

b) the column on which the inverted index has been created gets unloaded and loaded rather often.

In such a scenario it might be better to have the index also stored on disk.

Also, and that is a major reason, the index is immediately usable again after a fail-over to a standby node.

With SPS 7 the default behavior changed, so that the inverted indexes get stored to disk as well.

Hope that answers the question.

- Lars

[Former Member · Accepted solution]

Works here 🙂

Actually my reading of the RoE is that it recommends real names, rather than mandates them. I'm supposing this is their C-User?

Lars, if I understand correctly, the behavior was changed for Rev.70 but it doesn't work correctly, so has to be backed out until Rev.71?

John

[lbreddemann · Accepted solution]

Yep - now the RoE work here again, too.

And sure enough there are barely any strict rules in there, which is good.

I am just a fan of real name usage for active forum activity.

Back to the original topic: my understanding is that there is a bug in rev. 71 that could - under certain conditions, sometimes, not all the time, maybe not even on your system... you get the gist here - lead to rubbish being left in the inverted index structure.

If this inverted index is used then, well, clearly the good old garbage-in-garbage-out occurs and your result set contains wrong data.

Therefore, switch back to non-persisted inv. indexes for now and all most will be good.

Honestly, on my test system I couldn't easily create a test case where the different settings actually changed a great deal of the loading time. Looks to me as if this change of default behavior is rather neutral in most cases (except for this error of course ).

- Lars