Summary:
This should fix the issue of having a write followed by a read during
distributed execution. Any kind of merger of plans should behave like a
Cartesian with regards to planning the following ScanAll.
Reviewers: mtomic, mferencevic
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1624
Summary:
This diff splits single node and distributed storage from each other.
Currently all of the storage code is copied into two directories (one single
node, one distributed). The logic used in the storage implementation isn't
touched, it will be refactored in following diffs.
To clean the working directory after this diff you should execute:
```
rm database/state_delta.capnp
rm database/state_delta.hpp
rm storage/concurrent_id_mapper_rpc_messages.capnp
rm storage/concurrent_id_mapper_rpc_messages.hpp
```
Reviewers: teon.banek, buda, msantl
Reviewed By: teon.banek, msantl
Subscribers: teon.banek, pullbot
Differential Revision: https://phabricator.memgraph.io/D1625
Summary:
TODO:
~~1. Figure out how to propagate exceptions during lambda evaluation to master.~~
~~2. Make some more complicated test cases to see if everything is~~
~~sent over the network properly (lambdas depending on frame, evaluation context).~~
~~3. Support only `GraphView::OLD`.~~
4. [MAYBE] Send only parts of the frame necessary for lambda evaluation.
~~5. Fix EdgeType handling~~
--------------------
Serialize frame and send it in PrepareForExpand RPC
Move Lambda out of ExpandVariable
Send symbol table and filter lambda in CreateBfsSubcursor RPC
Evaluate filter lambda during the expansion
Send evaluation context in CreateBfsSubcursor RPC
Reviewers: teon.banek, msantl
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1600
Summary:
This should allow us to more easily decouple the code which should be
open sourced. Unfortunately, the downside of this approach is that we
cannot rely on virtual calls to dispatch the serialization to correct
type. Another downside is that members need to be publicly accessible
for serialization.
Reviewers: mtomic, msantl
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1596
Summary:
This diff changes the RPC layer to directly return `TResponse` to the user when
issuing a `Call<...>` RPC call. The call throws an exception on failure
(instead of the previous return `nullopt`).
All servers (network, RPC and distributed) are set to have explicit `Shutdown`
methods so that a controlled shutdown can always be performed. The object
destructors now have `CHECK`s to enforce that the `AwaitShutdown` methods were
called.
The distributed memgraph is changed that none of the binaries (master/workers)
crash when there is a communication failure. Instead, the whole cluster starts
a graceful shutdown when a persistent communication error is detected.
Transient errors are allowed during execution. The transaction that errored out
will be aborted on the whole cluster. The cluster state is managed using a new
Heartbeat RPC call.
Reviewers: buda, teon.banek, msantl
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1604
Summary:
Even though we may not need the results of a RPC, we should check that
it completed without error.
Reviewers: mferencevic, mtomic
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1610
Summary:
Knowledge of distributed operators is now removed from PlanPrinter
class. The DistributedOperatorVisitor and PlanPrinter are modified so
that they may be multiple inherited. This is done using virtual
inheritance of HierarchicalLogicalOperatorVisitor. Multiple inheritance
is used to derived a DistributedPlanPrinter which knows how to print
distributed in operators. This removes the dependency of single node
pretty printing on distributed.
Reviewers: msantl, mtomic
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1606
Summary:
Previous version of code was wrong in some cases. Example:
There's an edge e = (a)->(b). Node (a) belongs to worker 1, node (b) belongs to worker 0. Therefore, edge e belongs to worker 1.
When edge e was serialized on worker 0, address of node b was local and it would be globalized, but the worker id of edge (e) would be used, which is wrong.
Reviewers: teon.banek
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1599
Summary:
In a bunch of places `TypedValue` was used where `PropertyValue` should be. A lot of times it was only because `TypedValue` serialization code could be reused for `PropertyValue`, only without providing callbacks for `VERTEX`, `EDGE` and `PATH`. So first I wrote separate serialization code for `PropertyValue` and put it into storage folder. Then I fixed all the places where `TypedValue` was incorrectly used instead of `PropertyValue`. I also disabled implicit `TypedValue` to `PropertyValue` conversion in hopes of preventing misuse in the future.
After that, I wrote code for `VertexAccessor` and `EdgeAccessor` serialization and put it into `storage` folder because it was almost duplicated in distributed BFS and pull produce RPC messages. On the sender side, some subset of records (old or new or both) is serialized, and on the reciever side, records are deserialized and immediately put into transaction cache.
Then I rewrote the `TypedValue` serialization functions (`SaveCapnpTypedValue` and `LoadCapnpTypedValue`) to not take callbacks for `VERTEX`, `EDGE` and `PATH`, but use accessor serialization functions instead. That means that any code that wants to use `TypedValue` serialization must hold a reference to `GraphDbAccessor` and `DataManager`, so that should make clients reconsider if they really want to use `TypedValue` instead of `PropertyValue`.
Reviewers: teon.banek, msantl
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1598
Summary:
This is a simple change which modifies interface of
awesome_memgraph_functions to accept C-style pointer to array with
count. Doing things this way, allows us to easily try out different
allocation schemes for function arguments. In this diff, we are now
using stack allocation of arguments in a plain fixed size array. This is
done when the number of arguments is small. According to heaptrack, this
small change should yield noticeable improvements to heap usage.
Obviously, this doesn't solve the problem of heap allocations inside
TypedValue arguments themselves. These allocations appear when
std::string and std::vector is used inside TypedValue.
Micro benchmarks show that there is some performance improvement,
mostly around the limits of using array vs std::vector. The improvement is
more noticeable with multiple threads, due to primary gain being in avoiding
calls to memory allocation.
Reviewers: mtomic, msantl, mferencevic
Reviewed By: mferencevic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1581
Summary: A quick clean-up of user visible error messages. Tried to make them gramatically correct by capitalizing the first word in the sentence and putting a dot at the end.
Reviewers: teon.banek, buda, ipaljak
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1571
Summary:
Changed GRANT ROLE to SET ROLE. Now it is `SET ROLE FOR user TO ROLE` instead of `GRANT ROLE role TO user`. It makes more sense because our users can only have 1 role.
Changed REVOKE ROLE to CLEAR ROLE. Now it is `CLEAR ROLE FOR user` instead of `REVOKE ROLE role FOR user`. REVOKE ROLE would throw exception if user was not a member of role. CLEAR ROLE clears the role whatever it is. I find that the latter makes more sense combined with SET ROLE.
Changed `SHOW ROLE FOR USER user` to `SHOW ROLE FOR user`.
Changed `SHOW USERS FOR ROLE role` to `SHOW USERS FOR role`.
Reviewers: mferencevic, teon.banek, buda
Reviewed By: mferencevic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1572
Summary:
Visitor pattern's main issue is cyclical dependency between classes that
are visited and the visitor instance itself. We need to decouple this
dependency if we want to open source part of the code, namely
non-distributed part. This decoupling is achieved through the use of
`dynamic_cast` in distributed operators. Hopefully the solution is good
enough and doesn't cause performance issues. An alternative solution is
to build our own custom double dispatch solution, but that will
basically boil down to our implementation of runtime type information
and casts.
Note, this only decouples the distributed operators. If and when we
decide that other operators shouldn't be open sourced, the same
`dynamic_cast` pattern should be applied in them also.
Depends on D1563
Reviewers: mtomic, msantl, buda
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1566
Summary:
This is the first step in separating the implementation of distributed
features in operators. Following steps are:
* decoupling distributed visitors
* injecting distributed details in operator state
* minor cleanup or anything else that was overlooked
Reviewers: mtomic, msantl
Reviewed By: msantl
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1563
Summary:
The Kafka Python transform functionality uses a Python script to transform
incoming Kafka data into queries and parameters that are executed against the
database. When starting the Python transform script it is started in a
sandboxed environment so that it can't do harm to the host system or the
database.
Reviewers: msantl, teon.banek
Reviewed By: msantl
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1509
Summary:
Since we switched to Cap'n Proto serialization there's no need for
keeping boost around anymore.
Reviewers: mtomic, mferencevic, buda
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1515
Summary: first step for a cleaner query front end that might be easier to open source
Reviewers: teon.banek
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1511
Summary:
GraphDbAccessor is now constructed only through GraphDb. This allows the
concrete GraphDb to instantiate a concrete GraphDbAccessor. This allows
us to use virtual calls, so that the implementation may be kept
separate. The major downside of doing things this way is heap allocation
of GraphDbAccessor. In case it turns out to be a real performance
issues, another solution with pointer to static implementation may be
used.
InsertVertexIntoRemote is now a non-member function, which reduces
coupling. It made no sense for it to be member function because it used
only the public parts of GraphDbAccessor.
Reviewers: msantl, mtomic, mferencevic
Reviewed By: msantl
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1504
Summary:
This change, hopefully, simplifies the implementation of different kinds
of GraphDb. The pimpl idiom is now simplified by removing all of the
crazy inheritance. Implementations classes are just plain data stores,
without any methods. The interface classes now have a more flat
hierarchy:
```
GraphDb (pure interface)
|
+----+---------- DistributedGraphDb (pure interface)
| |
Single Node +-----+------+
| |
Master Worker
```
DistributedGraphDb is used as an intermediate interface for all the
things that should work only in distributed. Therefore, virtual calls
for distributed stuff have been removed from GraphDb. Some are exposed
via DistributedGraphDb, other's are only in concrete Master and Worker
classes. The code which relied on those virtual calls has been
refactored to either use DistributedGraphDb, take a pointer to what is
actually needed or use dynamic_cast. Obviously, dynamic_cast is a
temporary solution and should be replaced with another mechanism (e.g.
virtual call, or some other function pointer style).
The cost of the above change is some code duplication in constructors
and destructors of classes. This duplication has a lot of little tweaks
that make it hard to generalize, not to mention that virtual calls do
not work in constructor and destructor. If we really care about
generalizing this, we should think about abandoning RAII in favor of
constructor + Init method.
The next steps for splitting the dependencies that seem logical are:
1) Split GraphDbAccessor implementation, either via inheritance or
passing in an implementation pointer. GraphDbAccessor should then
only be created by a virtual call on GraphDb.
2) Split Interpreter implementation. Besides allowing single node
interpreter to exist without depending on distributed, this will
enable the planner and operators to be correctly separated.
Reviewers: msantl, mferencevic, ipaljak
Reviewed By: msantl
Subscribers: dgleich, pullbot
Differential Revision: https://phabricator.memgraph.io/D1493
Summary: This is a simple thing that enables us to use keywords as symbolic names. A bad thing is that planning will be done for queries which differ only in keyword case (e.g. "MATCH (x) RETURN x" and "match (x) return x"), but that should not be a significant performance impact as queries are done programmaticaly when high throughput is expected and we don't expect letter case change in that case.
Reviewers: buda, mferencevic, teon.banek
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1496
Summary:
First iteration in implementing kafka.
Currently, memgraph streams won't use the transform script provided in the
`CREATE STREAM` query.
There is a manual test that serves a POC purpose which we'll use to fully wire
kafka in memgraph.
Since streams need to download the script, I moved curl init from
telemetry.
Reviewers: teon.banek, mferencevic
Reviewed By: mferencevic
Subscribers: ipaljak, pullbot, buda
Differential Revision: https://phabricator.memgraph.io/D1491
