Press "Enter" to skip to content

Day: January 5, 2022

Improving Apache Flink Scheduler Performance

Zhilong Hong, et al, share some interesting results out of Apache Flink 1.14. Part one lays out the scene:

To estimate the effect of our optimizations, we conducted several experiments to compare the performance of Flink 1.12 (before the optimization) with Flink 1.14 (after the optimization). The job in our experiments contains two vertices connected with an all-to-all edge. The parallelisms of these vertices are both 10K. To make temporary deployment descriptors distributed via the blob server, we set the configuration blob.offload.minsize to 100 KiB (from default value 1 MiB). This configuration means that the blobs larger than the set value will be distributed via the blob server, and the size of deployment descriptors in our test job is about 270 KiB. The results of our experiments are illustrated below:

Part two explains their improvements:

In Flink 1.12, the ExecutionEdge class is used to store the information of connections between tasks. This means that for the all-to-all distribution pattern, there would be O(n2) ExecutionEdges, which would take up a lot of memory for large-scale jobs. For two JobVertices connected with an all-to-all edge and a parallelism of 10K, it would take more than 4 GiB memory to store 100M ExecutionEdges. Since there can be multiple all-to-all connections between vertices in production jobs, the amount of memory required would increase rapidly.

As we can see in Fig. 1, for two JobVertices connected with the all-to-all distribution pattern, all IntermediateResultPartitions produced by upstream ExecutionVertices are isomorphic, which means that the downstream ExecutionVertices they connect to are exactly the same. The downstream ExecutionVertices belonging to the same JobVertex are also isomorphic, as the upstream IntermediateResultPartitions they connect to are the same too. Since every JobEdge has exactly one distribution type, we can divide vertices and result partitions into groups according to the distribution type of the JobEdge.

Click through for a dive into the architecture.

Comments closed

Multiple Code Panes in R Studio

Tomaz Kastrun has good news for us:

On R studio home page, make sure to download the version 2021.09 Preview (as of writing of this blogpost, this is still in preview) and install this version on your client machine (supported windows machine, MacOS and Linux).

Once installation is completed, head to global options (Tools->Global options) and select Pane Layout. You will have a new set of buttons available (Add Column; Remove Column). With Add column an additional pane will be added to layout.

It’s not as convenient as the right-click –> “Split horizontally” or “Split vertically” that we get in tools like SSMS and VS Code, but I’m happy to see this change in R Studio.

Comments closed

Replication Error 20084 on SQL Server 2019

I ran into a weird issue:

Iwas helping out with a SQL Server upgrade recently, going from 2016 to 2019. We ran into a problem when trying to run replmerg.exe for a merge replication subscription. Specifically, we were getting error code 20084, which means that the replication process couldn’t connect to one of the instances. Interestingly, the process couldn’t connect to the local instance, and the failure was immediate—that is, within a couple of milliseconds. There was nothing in the management logs on either the distributor server or the subscriber server which indicated a problem. We were able to connect both sides together just fine—from the subscriber, we could connect to the distributor, and from the distributor, we could connect to the subscriber.

Click through for what error code 20084 typically means, as well as what turned out to be the problem here.

Comments closed

Swap an Identity Column with a Sequence

Greg Larsen looks at sequences:

Microsoft introduced the sequence number objects starting with SQL Server 2012. A sequence object generates sequence numbers based on starting and increment values, similar to an identity column, but it has additional features. Over time, you might find that the additional benefits of a sequence number have you wanting to replace an identity column with a sequence number. This article demonstrates two options for replacing an identity column with a sequence number.

When it came out, I thought I was going to use sequences a quite often. In practice, I’ve used it a few times, but IDENTITY is so much easier to type and I’ve rarely needed sequence generators. That said, the times I have needed them, I definitely appreciate their existence.

Comments closed

Power BI Model Documenter v2

Marc Lelijveld announces a new version of the Power BI Model Documenter:

Back in 2020, I released the first version of the Power BI external tool to document your Power BI data model. Since then, I wrote a lot about this topic, such as why adding descriptions to everything is important and various releases of the model documenter.

Users encountered various challenges with the initial release of the tool. I tried to help everyone to my best knowledge, but some issues kept coming back. Over the past period, I worked together with my colleague Ton Swart to solve all these challenges in a new updated version!

Read on to see what has changed.

Comments closed

Finding Substrings in a String with T-SQL

Kevin Wilkie avoids a regex:

Continuing on with our series from last time – see here if you somehow missed it – let’s have some more fun with the different functions we can use with strings.

This time, let’s focus on looking for different items we can use to find a string within a string.

With T-SQL not natively supporting regular expressions—though you can use a CLR module to do this—click through to see what Kevin uses.

Comments closed

2022 Data Professional Salary Survey

Brent Ozar wraps up another year of surveying:

Every year, I run a salary survey for folks in the database industry. This year, I was especially curious to see the results to find out whether salaries went up. Anecdotally, I’ve seen a lot of people jumping ship to new companies due to the Great Resignation – but what does the data actually show? Let’s find out.

Click through to grab a copy of the survey and get analyzing.

Comments closed

Serializing Clustered Columnstore Index Deletes

Aaron Bertrand hit a wall:

At Stack Overflow, we have some tables using clustered columnstore indexes, and these work great for the majority of our workload. But we recently came across a situation where “perfect storms” — multiple processes all trying to delete from the same CCI — would overwhelm the CPU as they all went widely parallel and fought to complete their operation. Here’s what it looked like in SolarWinds® SQL Sentry®:

It looks bad. Click through to understand why and what Aaron & co did to prevent this issue. I typically have used queue tables on the other end of columnstore indexes: as a method for ensuring that we insert 1024*1024 rows at a time. This was particularly important in the 2016 days, as we had a problem in which trickle-loading a columnstore index would cause massive numbers of rowgroups with dozens of rows, though that issue was subsequently fixed.

Comments closed