Store: Delta resize is not thread-safe about hyrise-v1

[grundprinzip]

yup, this is true :(

[schwald]

Could we use deque instead of the vectors in the delta?

[grundprinzip]

There are two things to consider here:

• reseve() in FixedLengthVector -> here a reallocate is called.... no guarantees
• resize() for the TX vectors in the store

No apparent solution comes to my mind that would allow fixing this...unfortunately

[schwald]

If we would use a deque as a container instead of vectors, this guarantees that references to elements stay valid even if the memory needs to be extended. However, iterators might get still invalidate which could be a problem...

http://stackoverflow.com/questions/913980/confusion-on-iterators-invalidation-in-deque
http://stackoverflow.com/questions/10373985/c-deque-when-iterators-are-invalidated

[bastih]

http://www.threadingbuildingblocks.org/docs/help/reference/containers_overview/concurrent_vector.htm ?

[grundprinzip]

This only solves the problem for the TX vectors, but not the problem for the actual storage. Iterators are not really used in our case. What we could do is to perform some stitching in the FLV. Like allocate a block and keep track of all used blocks etc... This will be a bit overhead but accetable as long as the block size is large enought.

[bastih]

Like allocate a block and keep track of all used blocks etc..

What exactly are you proposing?

I think we should distinguish between FixedStorage (-> as in ungrowable after construction, used in Main) and GrowableStorage (-> threadsafe growable and probably powered by a concurrent_vector-like structure, used in Delta).