Nebula Graph NebulaGraph for Beginners, 20 Jul 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Nebula for Beginners - Index Demystified (CN version & EN Subtitle)

Description

In this video, Siwei will talk you through the index in Nebula Graph and clarify when and how to use them. Hope you enjoy it. If you have any question, welcome to join our Slack channel: nebulagraph.slack.com
Tips: if you want to see English subtitle, please switch the subtitles to English in the settings.

A

Hey ~, hello, everyone, I'm hera! This is thought to.

B

Hi ~, hera, hi, ~, hello, everyone, I think, is very glad to meet you.

A

Is it a monday? Will.

A

This week we look at the index this one in figure, query inside very important function,.

A

To think of it, first, please tell you about the nebula graph inside the indexing.

A

B

First, we start from what it is talking about.

B

What is it?, we can see that it is a secondary index for the index attributes, rather than on the primary key to do.

B

It some information, such as a vertex, is the main key attribute vid and tag inside information.

B

That an index does,, it is only the information about the properties as an input.

B

Index it is used by only condition, attribute this situation to counter-check the.

B

Vid,, rather than in the process of expanding our example,, it is possible to filter attribute to it.

B

But that is the need to manually create the index,, so we are now this example. Lookup is a use of a query to the index.

B

We see it is that you want to find such a such a vertex belonging to tag1 and conditions of.

B

Its properties to meet column1 is greater than 32. Column2 is equal to the negative.

B

Of such a query,, this our index is a need of a query.

B

Secondly, it is worth noting that query in the index is only 2.0 inside the same type of support.

B

Such as the same type of vertex or a side of the same type.

B

More thing you should note is based on the index. The query is done in accordance with the matching left, the field.

B

Defined by the index, for a while, we will reveal more details.

B

In this respect, it is worth emphasizing here is what is not indexed?. We have said that the index is not used as the expansion of go of this.

B

Filter attribute,, which does not require the index followed by the index, is not coming to a query-based text-mode.

B

We had another function called a full-text index is used to do this.

B

Next scene, we simply look great help index designed nebula graph.

B

This will us to understand its capabilities,, its use, conditions and restrictions.

B

On the right is an architecture nebula graph, below is storaged inside each of.

B

Which represents a number of services entity.

B

Which uses the so-called local-index.

B

For the index of an instance of a rocksdb of space that native design, it means that our data.

B

Is one to one and index, index and point edge the existence of a local-storage inside.

B

Or inside a rocksdb such locality do it is to go to the local data of some benefits, such as for the so-called narrow thin-index index is the case then.

B

It can not get you across hosts to the request. Other properties.

B

By this point, other than the index.

B

There is consistency of this data and make it easier to achieve.

B

Our index look inside kv, design data and the index of.

B

The underlying storage,, the data inside it.

B

Kv design like pictures like this, in the type, the pid it's the type and index.

B

Type 2.0 inside it is a sub-point data type edge data.

B

Is partition is the id following.

B

Fragment of it is that we take is that here we only show point. The index format is similar to the side of.

B

The index, which is the point of this key it from left to right and then to assemble the first portion, is then first type.

B

This is an index that is then partition type id and the slice I -id is indexid own id index is. The next part is very important. Is the index-binary.

B

That all these column properties of this index, which is based on a binary way.

B

To save the binary from left to right and then inside this store, which explains that's: why.

B

Mentioned before us, our native index is left match.

B

Because it can go to your search in rocksdb inside.

B

And then we introduce the index read after write path. First of all, from the start of writing,.

B

It because you want to ensure that the data with the index, the consistency of the two parts of the data.

B

So that part of it is actually assigned to the following steps: first is to read the old data.

B

And then it going to build an old index key and use the build out key to delete the old index.

B

End it , and then inserted together new data and new index data.

B

So that this side, we note that it read the old data, it is a random read here- is actually at a cost of some performance.

B

But this is no way to avoid, because to ensure consistency needs to be done.

B

Is to write just read this path. We look at what is read.

B

First path, check expression is an expression over it,. We have to do it then.

B

Parses and checks example,. We will see this is a query with the previous example,. The same query.

B

Is found in the index column1 column2 under a tag1, this type.

B

That the next step is to find all valid index. We see for us tag1 index, which.

B

Has failed to comply with and relate to colume1 colume2.

B

It that the next step is to find the optimal index of our more than one index to meet this.

B

Kind of query that how to be more efficient as it currently is,, according to some of our established for example,, to query optimization rules in this case, which we colume2 = negative of this condition, is a.

B

Query such as equal it,, this proportion filter.

B

Query on the high probability, we will give priority to more equal conditions.

B

In advance farther, this rule was left in our class example, where.

B

I chose to second index is colume2 in front colume1. In such an index, 2.

B

That the next step is to get the final go inside the store that this step, because we are the so-called we do not place a global.

B

Local-Index distribution, so that it is.

B

Aware of our specific index in which, a few years, so we slice.

B

This is the first step to fanout to broadcast this to all the partiton.

B

Then there will first do an index prepared, ready to do this. It must first determine a few things.

B

We do that to the storage scan, scan a range of queries.

B

In fact, it will see us there are no return index is beyond our scope, if any,, that that this is a so-called after.

B

That,, if any,, that this is a so-called narrow index thin-index that case, we would also like to record, is.

B

Done after the real scan also have to go to a random read to read the data in which then read this property to construct this part of the return.

B

That this is optional,. If we already contains all this index return, then there is no.

B

More so this step,, that is all part of some basic introduction index of.

B

Good, thank you for your introduction.. When should we use the index?.

A

~, this is a good film is a good question.

B

Before hera, when we need it, the index can go back to.

B

That original index is a secondary index for the property, so it is only used by the query attribute.

B

This situation will require a vid of so when is it needed?

B

vid is when we do not know of such a query by the conditions you want to attribute.

B

To distinguish when we need to use a good.

B

Index, then, the next question is cited: nebula native and full-text cited differences. Where is it?.

A

There first, before the.

B

Full-Text index it is to support the properties of fuzzy, query text,, but the native index is left match. We talked.

B

Followed by native cited synchronous, write.

B

Us write path where introduced, the time of writing. It is to ensure consistency. There. Several steps of the operation of additional.

B

Full-Text index,, but it is an asynchronous write, and write it to build and index are not.

B

Part of the cluster nebula itself inside the computing power.

B

It is to achieve additional external elastic listener to do.

B

That last, if we just want a more detailed understanding of the full-text index,, then.

B

We suggest that you look at it official. This article describes in detail the information xiexie si to explain.

A

Our full-text index, an index that can affect the performance of queries, it?.

A

The impact on native index write performance is actually quite large,. We have already mentioned several steps before the newly introduced inside when talking about.

B

Writing,, especially the first step in the path of reading old data,, which is a random, read.

B

That its impact in fact, in the absence of the hit cache, is very large.

B

So we understand these all of these features and its format and its mechanism after.

B

Also help us to how to design and selection,, thereby reducing the impact on performance as much as possible.

B

Good questions and answers about this issue. The index on here.

A

We want to help,, if you have any doubts, welcome in the comments below the video.

A

Thank you, see the next issue. Bye.