Anybody know of any Graph DB's that support time series data?
Ideally we're looking for one that will scale well, and ideally use Cassandra or HBase as their persistent store.
Why would you want to do that? Best practice would be to store the dependency graph (in other words, the "Model" of the time series data) in a graphdb, but the actual time series in something more suited to that. Eg a KV store or a log-specific tool like Splunk...
See the KNMI (Dutch Weather Service) example for a case study: http://vimeopro.com/neo4j/graphconnect-europe-2015/video/128351859
Cheers!
Rik
One convenient way of doing that is to build a tree structure, with a common root, years as children, months as children of each year and down to the desired granularity.
At the end you attach the events nodes as leaf to this tree, and that give you the possibility to make many types of queries, from single point in time, to ranges, and also reverse, from the event to the timestamp...
Here is an example of this concept and an implementation within Neo4j
Related
I'm struggling to find a real world example on how to use google cloud dataflow combiners to run a common ETL tasl which aggregates records on multiple keys (e.g. Date, Location) and sums values over different measures (e.g. GrossValue, NetValue, Quantity). I can only find examples with a typical Key/Value (e.g. Day/Value) aggregation. Any hints on how this is done with the Python SDK would be appreciated.
I'm not 100% sure I understand your question. Do you have separate elements you are trying to join the data together for, in which case you may wish to use CoGroupByKey? Or does a single element have multiple fields?
Hope some of this info helps,
I would suggest looking at windowing, which will allow you to subdivide a PCollection according to the timestamps of its individual elements. If you want to see all the events for particular day this may be useful. Python examples of windowing. You may want to window across a days worth of data. This link is useful as well to understand how you can use GroupByKey in different ways,
Another option is to determine what date your elements belongs to, and use a group by key to key it with "[location][date][other]". You may need to do something like this if you want to join the data based on multiple fields.
See this GroupByKey example, but change the key to use your multiple fields concatenated.
Here is an example for reducing with a custom combiner. You can add logic here to do a custom aggregation for multiple different measurements.
We're using Neo4J and liking it. We do all sorts of graphy things in it. However, some of what we do is not graphy. For example, we keep a log of all changes to a certain type of node:
(n)-[:CHANGE]->(c1)-[:CHANGE]->(c2) etc etc
This list of changes can get to be 20 or 30 c1 nodes long. While it looks weird, I don't have a real problem with it. (Of course, I am smarter now, and since each :CHANGE relationship has a date in it, I could porcupine all the c1 nodes right from n. But whatever.)
But what if I wanted to store large amounts of text, or images. Is there a problem with storing large amounts of data in a node? I could use a different database for these things, but this just increases the skill set required to run the business. And of course, joining data in two disparate databases is always a PITA.
So I need to worry about storing large amounts of text in a single property? Do I need to avoid creating logs in the way I did above?
Creating linked lists of events, whatever they may be, is fine. It'd even say it is graphy! We use this approach a lot, in the case of ChangeFeed for something similar to what you're doing.
In Neo4j, such a linked list is actually better than a "porcupine", because you have to traverse all the relationships anyway, if you're looking for all changes. But in the linked list case, you don't have to look at properties to order them. In fact, the best approach is a hybrid approach, like the TimeTree.
As for storing large amounts of text or images, Neo4j is not the best place for it. Another database would be best, especially if the volume of these is large ( > hundreds of thousands).
But if you do want to store them in Neo4j, one thing to keep in mind is that Neo4j will load all the properties of a node/relationship once a single property is accessed. So in order to achieve good performance even with text/images in Neo4j, I would store them in a node of their own. That way, you can only load them if you really need them, but not during regular traversals.
For example:
CREATE (b:BlogPost {title:'Neo4j Rocks', author:'Tony Ennis', date:".."})-[:HAS_BODY]->(:BlogPostBody {content:'..'})
Say I am managing collectibles. I have thousands of baseball trading cards, thousands still of gaming cards (think Magic: the Gathering), and then thousands and thousands of doilies.
The part of me that's been steeped in relational databases for 20+ years is uncomfortable with the idea of thousands of Neo4J nodes floating out in space.
So I am inclined to gather them all with a node such as (:BASEBALL_CARDS), (:MTG_CARDS), and of course (:DOILIES). The idea is that these are singletons.
Now if I want all baseball cards that perhaps refer to a certain player, I could do something like:
(:BASEBALL_CARDS)-[GATHERS]->(:BASEBALL_CARD)-[:FEATURES]->(p:PLAYER {name: '...'})
It's very comforting to have the :BASEBALL_CARDS singleton, but does it do anything more than could be accomplished by indexing :BASEBALL_CARD?
(:BASEBALL_CARD)-[:FEATURES]->(p:PLAYER {name: '...'})
Is it best-practice to have thousands of free-ranging nodes?
One exceptional strong point of the graph database is the local query: the relationship lives in the instance, not in the type. A particular challenge (apart from modelling well) is determining the starting point of the local query (and keeping it local, i.e., avoiding path explosions). In Neo4j 1.x your One Node was a way to achieve a starting point for a certain kind of query. With 2.x and the introduction of labels, indexing :BaseballCard is the standard way to accomplish the same. If the purpose of that One Node is as a starting point for the kind of query in your example, then you are better off using a label index. A common problem in 1.x was that a node with an increasing number of relationships of the same type and direction eventually becomes a bottle neck for traversals. People started partitioning your One Node into A Paged Handful of Nodes, something like
(:BaseballCards)-[:GATHERS]->(:BaseballCards1to10000)-[:GATHERS]->(:BaseballCard)
The purpose of finding a starting point for the local query is often better served by labels, perhaps in combination with a basic, ordinary, local traversal, than by A Handful of Nodes. Then again, if it calms your nerves or satisfies your sense of the epic to have such a node, by all means have it. Because of the locality of queries, it will do you no harm.
In your example, however, neither the One Node nor an index on :BaseballCard would best serve as the starting point of the local query. The most particular pattern of interest is instead the name of the player. If you index (:Player) on name you will get the best starting point. The traversal across the one or handful* of [:FEATURES] relationships is very cheap and with a simple test on the other end for the :BaseballCard label, you are done. You could of course maintain the One Node for all players that share a name...
In my most humble opinion there is little need for discomfort. I do, however, want to affirm and commend your unease, in this one regard: that the graph is most powerful for connected data. The particular connection gathering the baseball cards doesn't seem to add new understanding or improve performance, but wherever there is disconnected data there is the potential for discovering exciting and meaningful patterns. Perhaps in the future the cards will be connected through patterns that signify their range of value, or the quality of their lamination, or a linked list of previous owners, or how well they work as conversations starters on a date. The absence of relationships is a call to find that One Missing Link that brings tremendous insight and value into your data.
* Handful, assuming that more than one baseball card features the same player, or some baseball players are also featured on cards of Magic: The Gathering. I'm illiterate in both domains, so I want to at least allow for the possibility.
It is ironic that you are concerned about nodes "floating out in space", when the whole idea behind graph DBs is making the connections between nodes a first class DB construct.
But I think your actual concern is that nodes do not "belong to a table" (in relational DB parlance). So, you would feel more comfortable in creating a special singleton node that in some sense takes the place of a table, from which you can access all the nodes that ought belong to that table.
A node label can be seen as the equivalent of a "table name". So, not only is there no need for you to also create a singleton "table node", doing so would be wasteful in DB resources, and complicate and slow down your queries. And neo4j can quickly access all the nodes with the same label.
There are several possible ways I can think of to store and then query temporal data in Neo4j. Looking at an example of being able to search for recurring events and any exceptions, I can see two possibilities:
One easy option would be to create a node for each occurrence of the event. Whilst easy to construct a cypher query to find all events on a day, in a range, etc, this could create a lot of unnecessary nodes. It would also make it very easy to change individual events times, locations etc, because there is already a node with the basic information.
The second option is to store the recurrence temporal pattern as a property of the event node. This would greatly reduce the number of nodes within the graph. When searching for events on a specific date or within a range, all nodes that meet the start/end date (plus any other) criteria could be returned to the client. It then boils down to iterating through the results to pluck out the subset who's temporal pattern gives a date within the search range, then comparing that to any exceptions and merging (or ignoring) the results as necessary (this could probably be partially achieved when pulling the initial result set as part of the query).
Whilst the second option is the one I would choose currently, it seems quite inefficient in that it processes the data twice, albeit a smaller subset the second time. Even a plugin to Neo4j would probably result in two passes through the data, but the processing would be done on the database server rather than the requesting client.
What I would like to know is whether it is possible to use Cypher or Neo4j to do this processing as part of the initial query?
Whilst I'm not 100% sure I understand you requirement, I'd have a look at this blog post, perhaps you'll find a bit of inspiration there: http://graphaware.com/neo4j/2014/08/20/graphaware-neo4j-timetree.html
We are working on a system where users can define their own nodes and connections, and can query them with arbitrary queries. A user can create a "branch" much like in SCM systems and later can merge back changes into the main graph.
Is it possible to create an efficient data model for that in Neo4j? What would be the best approach? Of course we don't want to duplicate all the graph data for every branch as we have several million nodes in the DB.
I have read Ian Robinson's excellent article on Time-Based Versioned Graphs and Tom Zeppenfeldt's alternative approach with Network versioning using relationnodes but unfortunately they are solving a different problem.
I Would love to know what you guys think, any thoughts appreciated.
I'm not sure what your experience level is. Any insight into that would be helpful.
It would be my guess that this system would rely heavily on tags on the nodes. maybe come up with 5-20 node types that are very broad, including the names and a few key properties. Then you could allow the users to select from those base categories and create their own spin-offs by adding tags.
Say you had your basic categories of (:Thing{Name:"",Place:""}) and (:Object{Category:"",Count:4})
Your users would have a drop-down or something with "Thing" and "Object". They'd select "Thing" for instance, and type a new label (Say "Cool"), values for "Name" and "Place", and add any custom properties (IsAwesome:True).
So now you've got a new node (:Thing:Cool{Name:"Rock",Place:"Here",IsAwesome:True}) Which allows you to query by broad categories or a users created categories. Hopefully this would keep each broad category to a proportional fraction of your overall node count.
Not sure if this is exactly what you're asking for. Good luck!
Hmm. While this isn't insane, think about the type of system you're replacing first. SQL. In SQL databases you wouldn't use branches because it's data storage. If you're trying to get data from multiple sources into one DB, I'd suggest exporting them all to CSV files and using a MERGE statement in cypher to bring them all into your DB at once.
This could manifest similar to branching by having each person run a script on their own copy of the DB when you merge that takes all the nodes and edges in their copy and puts them all into a CSV. IE
MATCH (n)-[:e]-(n2)
RETURN n,e,n2
Then comparing these CSV's as you pull them into your final DB to see what's already there from the other copies.
IMPORT CSV WITH HEADERS FROM "file:\\YourFile.CSV" AS file
MERGE (N:Node{Property1:file.Property1, Property2:file.Property2})
MERGE (N2:Node{Property1:file.Property1, Property2:file.Property2})
MERGE (N)-[E:Edge]-(N2)
This will work, as long as you're using node types that you already know about and each person isn't creating new data structures that you don't know about until the merge.