Summary:
During the following scenario:
- start a HA cluster with 3 machines
- find the leader and start sending queries
- SIGTERM the leader but leave other 2 machines untouched
The leader would be stuck in the shutdown phase.
This was happening because during the shutdown phase of the Bolt server, a
`graph_db_accessor` would try to commit a transaction after we've already shut
down Raft server. Raft, although not running, is still thinking it's in the
Leader mode. Tx Engine calls the `SafeToCommit` method to Commit transactions,
and ends up in an infinite loop.
Since Raft was shut down it won't handle any of the incoming RPCs and won't
change it's mode.
The fix here is to shut down the Bolt server before Raft, so we don't have any
pending commits once Raft is shut down.
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1853
Summary:
I've refactored the integration test for HA so we can reuse the common
parts like starting/stopping workers.
I've also added a test that triggers the log compaction and it checks that the
snapshot that has been transferred is the same as the origin one.
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: mferencevic, pullbot
Differential Revision: https://phabricator.memgraph.io/D1847
Summary:
This change splits mg-communication into mg-communication and
mg-comm-rpc. The main reason for doing this, is to make separation of
enterprise features from community Memgraph more clear.
Reviewers: mferencevic, msantl
Reviewed By: msantl
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1844
Summary:
RuleBasedPlanner now generates only the regular ScanAll operations, and
Filter operations are appended as soon as possible. The newly added
Rewrite step, takes this operator tree and replaces viable Filter &
ScanAll operators with appropriate ScanAllBy<Index> operator. This
change ought to simplify the behaviour of DistributedPlanner when that
stage is moved before the indexed lookup rewrite.
Showing unoptimized plan in interactive planner is also supported.
Reviewers: mtomic, llugovic
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1839
Summary:
All AST nodes had a member `uid_` that was used as a key in
`SymbolTable`. It is renamed to `symbol_pos_` and it appears only in
`Identifier`, `NamedExpression` and `Aggregation`, since only those types were
used in `SymbolTable`. SymbolGenerator is now responsible for creating symbols
in `SymbolTable` and assigning positions to AST nodes.
Cloning and serialization code is now simpler since there is no need to track
UIDs.
Reviewers: teon.banek
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1836
Summary:
In this part of log compaction for raft, I've implemented snapshooting
and snapshot recovery. I've also refactored the code a bit, so `RaftServer` now
has a pointer to the `GraphDb` and it can do some things by itself.
Log compaction requires some further work. Since snapshooting isn't synchronous
between peers, and each peer can work at their own pace, once we've compacted
the log so that the next log to be sent to peer `x` isn't available anymore, we
need to send the snapshot over the wire. This means that the next part will
contain the `InstallSnapshotRPC` and then maybe one more that will implement the
logic of sending `LogEntry` or the whole snapshot.
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1834
Summary:
`ReplicationLog` had a classic off-by-one bug. The `valid_prefix`
variable wasn't set properly.
This diff also includes a poor man's version of a HA client. This client
assumes that all the HA instances run on a single machine and that the
corresponding Bold endpoints have open ports ranging from `7687` to
`7687 + num_machines - 1`.
This should make it easeir to test certain things, ie. disk usage, P25.
This test revealed the bug with `ReplicationLog`
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1813
Summary:
Variable expansions cannot appear in merge patterns or after updates,
so they can only be planned with GraphState::OLD. Because of that, it makes
sense to remove GraphView parameter from them to reduce confusion.
Reviewers: teon.banek, llugovic
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1825
Summary:
Implement proper plan cloning using LCP instead of hacking it with
serialization.
depends on D1815
Reviewers: teon.banek, llugovic
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1816
Summary:
Once a leader loses it's leadership, in order to handle hanging
transactions, we reset the storage and the transaction engine.
This requires to re-apply all the commited entries from the log.
Once we add snapshot (log compaction) we would need to do that also.
One thing to have in mind is the `election_timeout_min` parameter. If it's set
too low it could trigger leader re-election too often.
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1822
Summary:
use newly added LCP functionality to get rid of manually written
`Clone` functions in AST.
depends on D1808
Reviewers: teon.banek, llugovic
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1815
Summary:
When there was an empty line comment starting right at the end of the
query, stripper wouldn't properly change state from `IN_LINE_COMMENT` to `OUT`
and it would return the wrong length (0) from `MatchWhitespaceAndComments`.
Because of that, the two slashes would be interpreted as division operators.
Query "RETURN 5;//" would be changed by stripper into "RETURN 5; / /" which
obviously can't be parsed;
Reviewers: teon.banek
Reviewed By: teon.banek
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1819
Summary:
It seems like the HA benchmark is flaky and unrelated builds fail
because of it. Lets disable it for now until we figure out what is happening.
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1814
Summary:
Rename Context to ExecutionContext and make it struct
Move ParsingContext to cypher_main_visitor.hpp
Reviewers: mtomic, llugovic
Reviewed By: llugovic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1810
Summary:
In order to get more consistent results, give the benchmark a certain
amount of time it is supposed to run and not the number of queries.
The resluts on my machine are as following:
```
duration 10.0004
executed_writes 25190
write_per_second 2518.91
duration 10.0005
executed_writes 25096
write_per_second 2509.48
duration 10.0004
executed_writes 23068
write_per_second 2306.7
duration 10.0006
executed_writes 26390
write_per_second 2638.84
duration 10.0008
executed_writes 26246
write_per_second 2624.38
duration 10.0006
executed_writes 24752
write_per_second 2475.06
duration 10.0027
executed_writes 24818
write_per_second 2481.14
duration 10.0032
executed_writes 25148
write_per_second 2513.99
duration 10.0009
executed_writes 25075
write_per_second 2507.28
duration 10.0008
executed_writes 25846
write_per_second 2584.4
duration 10.0006
executed_writes 25671
write_per_second 2566.96
duration 10.0025
executed_writes 25983
write_per_second 2597.65
```
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1812
Summary:
Identifiers would be printed correctly if they were printed as part of a larger
`Expression` (so the visitor did the printing). However, if `PrintObject` was
called with an `Identifier *` then it would delegate to the template overload
and just print the pointer.
Reviewers: mtomic, teon.banek
Reviewed By: mtomic
Subscribers: mferencevic, pullbot
Differential Revision: https://phabricator.memgraph.io/D1802
Summary:
This diff removes the need for a database when parsing a query and
creating an Ast. Instead of storing storage::{Label,Property,EdgeType}
in Ast nodes, we store the name and an index into all of the names. This
allows for easy creation of a map from {Label,Property,EdgeType} index
into the concrete storage type. Obviously, this comes with a performance
penalty during execution, but it should be minor. The upside is that the
query/frontend minimally depends on storage (PropertyValue), which makes
writing tests easier as well as running them a lot faster (there is no
database setup). This is most noticeable in the ast_serialization test
which took a long time due to start up of a distributed database.
Reviewers: mtomic, llugovic
Reviewed By: mtomic
Subscribers: mferencevic, pullbot
Differential Revision: https://phabricator.memgraph.io/D1774
Summary:
Each `raft::LogEntry` is now persisted under its own key in our `KVStore`. Locally running our HA feature benchmark yields the following results:
```
duration 23.7
executed_writes: 15000
write_per_second: 632.888
```
This represents about 5x increase in throughput.
Reviewers: msantl
Reviewed By: msantl
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1799
Summary:
A simple benchmark that starts a HA cluster with 3 machines.
The benchmark issues only `CREATE (:Node)` queries.
Local results (debug build), for this raft config, are:
```
duration 4.26899
executed_writes 300
write_per_second 70.2743
```
Reviewers: ipaljak, mferencevic
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1798
Summary:
`Edges` class didn't properly filter edges when given both edge type
and destination arguments, which causes weird bugs in query execution (wrong
edge getting matched in `Expand` with existing node flag set). This is probably
because we didn't support using both filters at the same time, but the API and
documentation for `VertexAccessor::in` and `VertexAccessor::out` functions
didn't reflect that.
Reviewers: teon.banek, msantl
Reviewed By: msantl
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1796
Summary: Run the basic test with two cluster sizes, 3 and 5.
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1794
Summary:
Created a new integration test for Raft protocol.
The tests iterates through the Raft cluster and does the following:
* kill machine `X`
* execute a query
* bring `X` back to life
The first step is to insert a vertex in the cluster, and last step is to check
if the cluster has all the data.
I also edited some of the raft core files because this test surafaced some bugs.
The `tester` binary is a hacked version of the HA client and so are the parts in
the code that refuse to execute a query is the machine is not in `Leader` mode.o
Those parts will go away once we have a proper HA client.
I've run the `runner.py` for a while (215 times)
```
while ./runner.py &> log.txt; do echo -n "."; done
```
and it didn't break.
Reviewers: ipaljak, mferencevic
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1788
Summary: This simple function is required by the Tensorflow integration so that Memgraph can always return regular matrices of desired size.
Reviewers: teon.banek, mtomic, dsantl
Reviewed By: mtomic, dsantl
Subscribers: mferencevic, pullbot
Differential Revision: https://phabricator.memgraph.io/D1783
Summary:
This (almost) removes the dependency of operators on NodeAtom and
EdgeAtom. Only EdgeAtom::Direction is needed. The change was done as the
initial step of removing dependency on storage from Ast. Additionally,
it makes sense for LogicalOperator to only depend on Expression classes.
Reviewers: mtomic, llugovic
Reviewed By: llugovic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1779
Summary:
TransactionReplicator replicates transactions on follower machines in
HA memgraph. Our DB accessor API doesn't provide us with the functionality to
begin transactions with non-increasing ids. This is why the
`TransactionReplicator` uses a internal map that maps tx ids from the leader
node to transactions on the follower node (whose id doesn't have to match the
leaders tx id).
If the leader has the following transaction timeline:
```
L
tx1
|
| tx2
| |
| |
| |
| |
| |
| tx2
|
|
|
|
tx1
```
`tx2` will commit first and will be replicated. When applying `tx2` on follower
nodes, they will start a new transaction with tx id `1`. When `tx1` starts
replicating, followers will start a new transaction with tx id `2`. And this is
wehre `TransactionReplicator` kicks in.
Reviewers: ipaljak
Reviewed By: ipaljak
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1775
Summary:
This is just the first diff that tries to wire the raft protocol into
memgraph.
In this diff I'm introducing transaction engine reset functionality. I also
introduced `RaftInterface` which should be used wherever someone wants to access
Raft from Memgraph.
For design decisions see the feature spec.
Reviewers: ipaljak, teon.banek
Reviewed By: ipaljak
Subscribers: pullbot, teon.banek
Differential Revision: https://phabricator.memgraph.io/D1758
Summary:
With quite frequent changes of serialization backend, it's getting
really painful updating the C++ implementation. This diff defines the
Symbol class via LCP, so that the C++ serialization is generated instead
of written by hand.
Reviewers: mtomic, llugovic
Reviewed By: mtomic
Subscribers: pullbot
Differential Revision: https://phabricator.memgraph.io/D1741
