Performance after Adding Headless Modus

[drfho]

Ref: 0e13c11

Exactly this change-set massivly reduces the performance of attribute listings like in the "Meta-Attribute":
/content/metaobj_manager/manage_metas?lang=ger
The creation of the table usually tales 0.2s, now it takes a least 12s

The latency is due to the modification in _cachemanager.py
0e13c11#diff-93d213afd35022330e5586bb4981eca832e4420321cdcd98646286dc2dd11c86

If this change is revoked, the performance gets to expected speed, about.

Actualy when iterating the metadict there will be an intense ping-pong of fetchReqBuff() and storeReqBuff() which costs a lot of time. Now, because self=content the request will be alsways created new by standard.create_headless_http_request(). In fact, this is not needed here at all.

So we need a solution to avoid any redundant creation of the REQUEST-Object, that is just needed for headless requests and not ordinary usage of ZMS. Maybe as global var the is just created once on startup?

[cmeier76]

@drfho Maybe we should create the fallback REQUEST object before store/fetchReqBuff and pass it as parameter...

...I will check this ASAP.

Meanwhile you can revert the both destructive calls in main branch - we do not rely on this at the moment.

[cmeier76]

@drfho Your fix ac35c27 does work for the CMSAPI with object-relational mapping from ZODB to PostgreSQL as well - with a performance boost by factor 3. Thank you very much for spotting this issue!

[cmeier76]

Fixed by #322