Dynamic allocation memory (part 1) #120
Conversation
taitov
force-pushed
the
dyn-acl
branch
2 times, most recently
from
7e36d48 to
d2a5e84
Compare
taitov
force-pushed
the
dyn-acl
branch
2 times, most recently
from
9812d65 to
057cefc
Compare
taitov
changed the title
taitov
force-pushed
the
dyn-acl
branch
2 times, most recently
from
e7e9c63 to
1f17057
Compare
In this PR (part 1) new options is useless =( |
| stats.extended_chunks_size = std::max((uint64_t)object_type::extended_chunks_size_min, | ||
| values.size() / 16); | ||
|
|
||
| for (;;) |
There was a problem hiding this comment.
So you just double the size of a LPM tree in case of no memory error?
There was a problem hiding this comment.
No. Double size only if not enough chunks on update
| } | ||
|
|
||
| eResult result = pointer->fill(stats, values); | ||
| if (result != eResult::success) |
There was a problem hiding this comment.
Do we really want to double the size of a hashtable in case of ANY error?
There was a problem hiding this comment.
fill can return success, or not enough memory
|
|
||
| for (;;) | ||
| { | ||
| pointer = memory_manager->create<object_type>(name.data(), |
There was a problem hiding this comment.
I think, extended_chunks_size should be encapsulated into the tree but not the updater
There was a problem hiding this comment.
The idea was that worker and lookup is not required extended_chunks_size
| socket_id, | ||
| size); | ||
|
|
||
| void* pointer = rte_malloc_socket(nullptr, |
There was a problem hiding this comment.
I do not like the rte_malloc here as it could be fragile, for instance if we are near the overall allocation limits.
There was a problem hiding this comment.
At minimum a complex investigation in rte_malloc behavior should be done before its application for dynamic allocation purposes.
Personally, I am unsure that allocation sequences like A16 A16, A40 F16 F16 A40 (allocate 16GB twice, then reallocate the first one from 16 to 40 and then the second one 16 to 40) would be handled well.
There was a problem hiding this comment.
We will not make things worse in this place
allocate memory on update for some acl structs.
allocate memory on update for some acl structs
#58