perf: improve performance of bulk child entity updates and avoid hitting the AQL limit of 500 nesting limit doing it

[Yogu]

perf: improve performance of bulk child entity updates and avoid hitting the AQL limit of 500 nesting limit doing it

Child entity updates generate a lot of AQL because each entity can have different kinds of updates, and some of them refer to old values. This is still the case after the optimizations.

However, previously, we performed multiple child entity updates by nesting conditional expressions like this:

items = items.map(item =>
  item.id == 2 ? update2(item) : (item.id == 1 ? update1(item) : item)
);

This generated one level of nesting per update. ArangoDB 3.11 now limits the AQL nesting to 500, which means we would be limited to a little under 500 item updates in one mutation. In addition, a few hundred updates had realy bad performance.

Now, we convert the list into a dictionary (id -> entity), so we can more efficiently look up items and construct the new list.

This optimization is only applied after a threshold of 3 (configurable) because it involves some steps, and doing a single or two updates is probably still faster with the conditionals.

[mfusser]

to me this reads as "the threshold that needs to be reached to activate dict-strategy is 2. So it is active for 2" which is not the case.

[mfusser]

nevermind, it is the case. I just got confused because the default is supposed to be 3 and this is a file called "default context"

[Yogu]

yeah I initially had the (real) default at 5 and wanted to reduce it for the tests so we have more chances to trigger it, but I then dialed down the real default to 3. No idea what a good value could be yet.