Documentation

# HStreamDB release notes

# v0.9.0 [2022-07-29]

# HStreamDB

# Highlights

# Shards in Streams

We have extended the sharding model in v0.8, which provides direct access and management of the underlying shards of a stream, allowing a finer-grained control of data distribution and stream scaling. Each shard will be assigned a range of hashes in the stream, and every record whose hash of partitionKey falls in the range will be stored in that shard.

Currently, HStreamDB supports:

  • set the initial number of shards when creating a stream
  • distribute written records among shards of the stream with partitionKeys
  • direct access to records from any shard of the specified position
  • check the shards and their key range in a stream

In future releases, HStreamDB will support dynamic scaling of streams through shard splitting and merging

# HStream IO

HStream IO is the built-in data integration framework for HStreamDB, composed of source connectors, sink connectors and the IO runtime. It allows interconnection with various external systems and empowers more instantaneous unleashing of the value of data with the facilitation of efficient data flow throughout the data stack.

In particular, this release provides connectors listed below:

You can refer to the documentation to learn more about HStream IO.

# New Stream Processing Engine

We have re-implemented the stream processing engine in an interactive and differential style, which reduces the latency and improves the throughput magnificently. The new engine also supports multi-way join, sub-queries, and more general materialized views.

The feature is still experimental. For try-outs, please refer to the SQL guides.

# Gossip-based HServer Clusters

We refactor the hserver cluster with gossip-based membership and failure detection based on SWIM (opens new window), replacing the ZooKeeper-based implementation in the previous version. The new mechanism will improve the scalability of the cluster and as well as reduce dependencies on external systems.

# Advertised Listeners

The deployment and usage in production could involve a complex network setting. For example, if the server cluster is hosted internally, it would require an external IP address for clients to connect to the cluster. The use of docker and cloud-hosting can make the situation even more complicated. To ensure that clients from different networks can interact with the cluster, HStreamDB v0.9 provides configurations for advertised listeners. With advertised listeners configured, servers can return the corresponding address for different clients, according to the port to which the client sent the request.

# Improved HStream CLI

To make CLI more unified and more straightforward, we have migrated the old HStream SQL Shell and some other node management functionalities to the new HStream CLI. HStream CLI currently supports operations such as starting an interacting SQL shell, sending bootstrap initiation and checking server node status. You can refer to the CLI documentation for details.

# Monitoring with Grafana

We provide a basic monitoring solution based on Prometheus and Grafana. Metrics collected by HStreamDB will be stored in Prometheus by the exporter and displayed on the Grafana board. For details, refer to the documentation.

# Deployment on K8s with Helm

We provide a helm chart to support deploying HStreamDB on k8s using Helm. You can refer to the documentation for details.

# Java Client

The Java Client v0.9.0 (opens new window) has been released, with support for HStreamDB v0.9.

# Golang Client

The Go Client v0.2.0 (opens new window) has been released, with support for HStreamDB v0.9.

# Python Client

The Python Client v0.2.0 (opens new window) has been released, with support for HStreamDB v0.9.

# v0.8.0 [2022-04-29]

# HServer

# New Features

  • Add mutual TLS support
  • Add maxUnackedRecords option in Subscription: The option controls the maximum number of unacknowledged records allowed. When the amount of unacked records reaches the maximum setting, the server will stop sending records to consumers, which can avoid the accumulation of unacked records impacting the server's and consumers' performance. We suggest users adjust the option based on the consumption performance of their application.
  • Add backlogDuration option in Streams: the option determines how long HStreamDB will store the data in the stream. The data will be deleted and become inaccessible when it exceeds the time set.
  • Add maxRecordSize option in Streams: Users can use the option to control the maximum size of a record batch in the stream when creating a stream. If the record size exceeds the value, the server will return an error.
  • Add more metrics for HStream Server.
  • Add compression configuration for HStream Server.

# Enhancements

  • [breaking changes] Simplify protocol, refactored codes and improve the performance of the subscription
  • Optimise the implementation and improve the performance of resending
  • Improve the reading performance for the HStrore client.
  • Improve how duplicated acknowledges are handled in the subscription
  • Improve subscription deletion
  • Improve stream deletion
  • Improve the consistent hashing algorithm of the cluster
  • Improve the handling of internal exceptions for the HStream Server
  • Optimise the setup steps of the server
  • Improve the implementation of the stats module

# Bug fixes

  • Fix several memory leaks caused by grpc-haskell
  • Fix several zookeeper client issues
  • Fix the problem that the checkpoint store already exists during server startup
  • Fix the inconsistent handling of the default key during the lookupStream process
  • Fix the problem of stream writing error when the initialisation of hstore loggroup is incompleted
  • Fix the problem that hstore client writes incorrect data
  • Fix an error in allocating to idle consumers on subscriptions
  • Fix the memory allocation problem of hstore client's appendBatchBS function
  • Fix the problem of losing retransmitted data due to the unavailability of the original consumer
  • Fix the problem of data distribution caused by wrong workload sorting

# Java Client

# New Features

  • Add TLS support
  • Add FlowControlSetting setting for BufferedProducer
  • Add maxUnackedRecords setting for subscription
  • Add backlogDurantion setting for stream
  • Add force delete support for subscription
  • Add force delete support for stream

# Enhancements

  • [Breaking change] Improve RecordId as opaque String
  • Improve the performance of BufferedProducer
  • Improve Responder with batched acknowledges for better performance
  • Improve BufferedProducerBuilder to use BatchSetting with unified recordCountLimit, bytesCountLimit, ageLimit settings
  • Improve the description of API in javadoc

# Bug fixes

  • Fix streamingFetch is not canceled when Consumer is closed
  • Fix missing handling for grpc exceptions in Consumer
  • Fix the incorrect computation of accumulated record size in BufferedProducer

# Go Client

# Admin Server

# Tools

  • Add bench tools (opens new window)
  • [dev-deploy] Support limiting resources of containers
  • [dev-deploy] Add configuration to restart containers
  • [dev-deploy] Support uploading all configuration files in deploying
  • [dev-deploy] Support deployments with Prometheus Integration

# v0.7.0 [2022-01-28]

# Features

# Add transparent sharding support

HStreamDB has already supported the storage and management of large-scale data streams. With the newly added cluster support in the last release, we decided to improve a single stream's scalability and reading/writing performance with a transparent sharding strategy. In HStreamDB v0.7, every stream is spread across multiple server nodes, but it appears to users that a stream with partitions is managed as an entity. Therefore, users do not need to specify the number of shards or any sharding logic in advance.

In the current implementation, each record in a stream should contain an ordering key to specify a logical partition, and the HStream server will be responsible for mapping these logical partitions to physical partitions when storing data.

# Redesign load balancing with the consistent hashing algorithm

We have adapted our load balancing with a consistent hashing algorithm in this new release. Both write and read requests are currently allocated by the ordering key of the record carried in the request.

In the previous release, our load balancing was based on the hardware usage of the nodes. The main problem with this was that it relied heavily on a leader node to collect it. At the same time, this policy requires the node to communicate with the leader to obtain the allocation results. Overall the past implementation was too complex and inefficient. Therefore, we have re-implemented the load balancer, which simplifies the core algorithm and copes well with redistribution when cluster members change.

# Add HStream admin tool

We have provided a new admin tool to facilitate the maintenance and management of HStreamDB. HAdmin can be used to monitor and manage the various resources of HStreamDB, including Stream, Subscription and Server nodes. The HStream Metrics, previously embedded in the HStream SQL Shell, have been migrated to the new HAdmin. In short, HAdmin is for HStreamDB operators, and SQL Shell is for HStreamDB end-users.

# Deployment and usage

To make use of HStreamDB v0.7, please use hstreamdb-java v0.7.0 (opens new window) and above

# v0.6.0 [2021-11-04]

# Features

# Add HServer cluster support

As a cloud-native distributed streaming database, HStreamDB has adopted a separate architecture for computing and storage from the beginning of design, to support the independent horizontal expansion of the computing layer and storage layer. In the previous version of HStreamDB, the storage layer HStore already has the ability to scale horizontally. In this release, the computing layer HServer will also support the cluster mode so that the HServer node of the computing layer can be expanded according to the client request and the scale of the computing task.

HStreamDB's computing node HServer is designed to be stateless as a whole, so it is very suitable for rapid horizontal expansion. The HServer cluster mode of v0.6 mainly includes the following features:

  • Automatic node health detection and failure recovery
  • Scheduling and balancing client requests or computing tasks according to the node load conditions
  • Support dynamic joining and exiting of nodes

# Add shared-subscription mode

In the previous version, one subscription only allowed one client to consume simultaneously, which limited the client's consumption capacity in the scenarios with a large amount of data. Therefore, in order to support the expansion of the client's consumption capacity, HStreamDB v0.6 adds a shared-subscription mode, which allows multiple clients to consume in parallel on one subscription.

All consumers included in the same subscription form a Consumer Group, and HServer will distribute data to multiple consumers in the consumer group through a round-robin manner. The consumer group members can be dynamically changed at any time, and the client can join or exit the current consumer group at any time.

HStreamDB currently supports the "at least once" consumption semantics. After the client consumes each data, it needs to reply to the ACK. If the Ack of a certain piece of data is not received within the timeout, HServer will automatically re-deliver the data to the available consumers.

Members in the same consumer group share the consumption progress. HStream will maintain the consumption progress according to the condition of the client's Ack. The client can resume consumption from the previous location at any time.

It should be noted that the order of data is not maintained in the shared subscription mode of v0.6. Subsequent shared subscriptions will support a key-based distribution mode, which can support the orderly delivery of data with the same key.

# Add statistical function

HStreamDB v0.6 also adds a basic data statistics function to support the statistics of key indicators such as stream write rate and consumption rate. Users can view the corresponding statistical indicators through HStream CLI, as shown in the figure below.

# Add REST API for data writing

HStreamDB v0.6 adds a REST API for writing data to HStreamDB.

←

编辑此页
本页导航