README
¶
kateway
A fully-managed real-time secure and reliable geo-replicated RESTful Cloud Pub/Sub streaming message/job service.
_/ _/ _/
_/ _/ _/_/_/ _/_/_/_/ _/_/ _/ _/ _/ _/_/_/ _/ _/
_/_/ _/ _/ _/ _/_/_/_/ _/ _/ _/ _/ _/ _/ _/
_/ _/ _/ _/ _/ _/ _/ _/ _/ _/ _/ _/ _/ _/
_/ _/ _/_/_/ _/_/ _/_/_/ _/ _/ _/_/_/ _/_/_/
_/
Alternatives
- Google Cloud Pub/Sub
- Amazon kenesis/SQS
- kenesis is aws kafka
- SQS is aws beanstalkd
- Azure EventHub
- Yahoo Pulsar
- IBM Bluemix Message Hub
- uber Cherami
- qcloud CMQ
- aliyun MNS
- order not guaranteed
- backtracking not supported
- misc
- pubnub
- pusher
- firebase
- parse
- pubsubhubbub
- aliyun ONS
Features
- REST API
- http/https/websocket/http2 interface for Pub/Sub
- Support both FIFO and Schedulable queue
- Flexible delivery options
- Both push- and pull-style subscriptions supported
- Communication can be
- one-to-many (fan-out)
- many-to-one (fan-in)
- many-to-many
- Systemic Quality Requirements
- Performance & Throughput
-
100K msg/sec delivery on a single host without batch
- fully benchmark tested and profiler'ed
-
- Scalability
- scales to 1M msg/sec
- elastic scales
- Latency
- < 1s delivery
- Availability
- Graceful shutdown without downtime
- Long polling
- Graceful Degrade
- throttle
- circuit breaker
- hinted handoff
- Performance & Throughput
- Fully-managed
- Discovery
- Create versioned topics, subscribe to topics
- Rich real-time tagged metrics, fully-functional dashboard and alarming
- Easy trouble shooting
- Controlled GC
- Visualize message flow
- Managed integration service via Webhooks
- Hot configurable
- Multi-tenant
- Authentication
- Authorization
- Quotas
- Optional hardware isolation
- REST API for provisioning, admin and stats
- swagger documentation
- Mirror across data centers
- Replicated storage and guaranteed at-least-once message delivery
- Functional Features
- schedulable message
- server side message filter by tag
- managed message routing
- avro based message schema registration and versioning
- retry|dead queue
- sub in batch
- message backtracking
- hot dryrun topic
- multi-tenant metrics
- self-servicable topic scaling and message rentention SLA
- user can check sub status/lag
- topic owners can check subscribers and their status
- configurable lag alerting
- Enables sophisticated streaming data processing
- Load balancer friendly
- Quotas and rate limit, QoS
- Flow control: Dynamic rate limiting
- Encryption of all message data on the wire
Common scenarios
- Balancing workloads in network clusters
- Implementing asynchronous workflows
- Distributing event notifications
- Refreshing distributed caches
- Logging to multiple systems
- Data streaming from various processes or devices
- Reliability improvement
APIs
Pub
POST /v1/msgs/:topic/:ver
POST /v1/ws/msgs/:topic/:ver
POST /v1/jobs/:topic/:ver
DELETE /v1/jobs/:topic/:ver
Sub
GET /v1/msgs/:appid/:topic/:ver
GET /v1/ws/msgs/:appid/:topic/:ver
POST /v1/shadow/:appid/:topic/:ver/:group
DELETE /v1/groups/:appid/:topic/:ver/:group
GET /v1/subd/:topic/:ver
GET /v1/status/:appid/:topic/:ver
The Big Picture
+-----------+
| VirtualIP |
+-----------+
|
+--------------+ Alert SOS Dashboard
| | | | | gk
+-------------------------------------------------------------------------------------------+
| | | | | | |
| +----------+ +----------+ | | | |
| | ehaproxy | | ehaproxy | | V | |
| +----------+ +----------+ | | | |
| | | | discovery | | | |
| +----------------+ | +-------------+ |
| | LB | | |
| | | +--------------------+ +--------+ |
| | +---| | election | | watch |
| |keepalive | zookeeper ensemble |-----------| kguard |-------------+ |
| | +---| | | | aggragator | |
| | | +--------------------+ +--------+ | |
| | +-------+ | | |
| | | registry | orchestration | | +- Pub
| +---------------+ |-----------+ +---------+ | | REST |
| | | | | | kateway |--|----|------|
| +---------+ +---------+ +--------+ +--------+ +---------+ | | |
| | kateway | | kateway | | actord | | actord | | | +- Sub
| +---------+ +---------+ +--------+ +--------+ | |
| | hh | | hh | | executor | |
| +----+ +----+ +--------------+ | |
| | | | | |
| | +---------+ +---------+ push | |
| +--------+ | JobTube | | Webhook |------------>-----------------|----|---Endpoints
| | | +---------+ +---------+ | |
| | | | scheduler | sub | |
| auth | |job WAL | dispatch | | |
| +------+ +---------------------------+------------------+ | |
| | | tenant shard | pubsub | flush | |
| +----------+ +---------+ +-------+ +------+ | |
| | auth DB | | DB Farm | | kafka | | TSDB | | |
| +----------+ +---------+ +-------+ +------+ | |
| | | | | | |
| | +----------------------------------------------+ | |
| | | | |
| | +-------------------------------------------+ |
| | |
| | zone |
+-------------------------------------------------------------------------------------------+
|
WebConsole
Management
GET /alive
GET /v1/status
GET /v1/clusters
GET /v1/clients
GET /v1/partitions/:cluster/:appid/:topic/:ver
POST /v1/topics/:cluster/:appid/:topic/:ver
DELETE /v1/counter/:name
FAQ
-
why named kateway?
Admittedly, it is not a good name. Just short for kafka gateway
-
how to batch messages in Pub?
It is http client's job to put the variant length data into json array
-
how to consume multiple messages in Sub?
add param
batchwhen Sub. kateway uses chunked transfer encoding and client MUST use TLV to decode. -
http header size limit?
4KB
-
what is limit of a pub message in size?
1 ~ 512KB
-
what is limit of a job message in size?
1 ~ 16KB
-
if sub with no arriving message, how long do client get http 204?
30s
Dependencies
- github.com/samuel/go-zookeeper
35a5e3bf880ea78fbebad4c7f349d4852c2f75a6
- github.com/Shopify/sarama
d37c73f2b2bce85f7fa16b6a550d26c5372892ef
TODO
-
kguard watch GC for zk/kafka
-
tag move from body to key, only store hash(tag)
-
kw id replaced by zk sequence
-
cleanup of the idle->active pub clients during shutdown
-
sub status display raw kafka offset status
-
mirror, when destination dies stop consuming
-
tagged metrics
-
Content-Length if batch pub
-
haproxy metrics export to InfluxDB http://user:pass@localhost:10890/stats?stats;csv
-
raft consensus
-
hh should use AsyncPub method instead of SyncPub
-
hh index
-
integration: bad Pub/Sub rate limit
- Pub
- Sub
-
when consumer group decision didn't change, refuse to rebalance
-
Pub pool will create up to 100 * (count(topic) + count(partition)) goroutines
-
mysql slave of manager to data sync with manager in memory
-
kguard apperr.go Line:128 restart if kafka conn broken
-
kguard watch for same group consuming multiple topics
-
when startup, hh Empty?
-
man /v1/clusters director of cluster distribution
-
bug: /Users/funky/gopkg/src/github.com/samuel/go-zookeeper/zk/conn.go 511 panic: non-positive interval for NewTicker
-
go over all zk watchers, handle expire
- ehaproxy
- kateway registry
- actord
- kguard
-
ehaproxy sometimes haproxy process didn't exit normally
-
bug: when shutdown, consumer group is not sync'ed with sub server stop
-
1 consumer group can only Sub 1(not 2 or more) topic
- lessen the thundering hurd because rebalance when group members change
- better manageable
-
when delete a topic, remove its all consumer groups
-
bug: kateway gone, but kguard kateway.pubsub.fail didn't notice
-
zk session expires and ephemeral znodes?
-
IO load balance for hinted handoff
-
metrics for hh
-
swagger
-
reject Sub when the group is being used for another topic
-
pause a webhook
-
kafka timeout 1ms can work with hinted handoff
-
DFS/Kahn webhook dead loop detection
-
job
- pause/resume a job
- job state machine
- partition table?
-
deregister before web listener closed
-
github.com/funkygao/go-metrics/sample.go:151 heap
-
failure tolerance
- pub breaker
- migration partition from a to b
- what if a broker killed
-
async pub mem pool
-
sub: what if broker moved
-
topic name obfuscation
-
sub with delayed ack
- StatusNotModified
- what if rebalanced, and ack buffered p/o
-
pub/sub a disabled topic, discard?
-
features confirm
- delayed job
- bury
- msg tag
- avro schema registry
-
fetchShadowQueueRecords enable
-
Plugins
- authentication and authorization
- transform
- hooks
-
check hack pkg
-
https, outer ip must https
-
Update to glibc 2.20 or higher
-
compress
- gzip sub response
- Pub HTTP POST Request compress
- compression in kafka
Documentation
¶
There is no documentation for this package.
Source Files
Directories
¶
| Path | Synopsis |
|---|---|
|
api
|
|
|
Pub/Sub benchmark webhookd is a webhook endpoint that is used to demo kateway webhook feature.
|
Pub/Sub benchmark webhookd is a webhook endpoint that is used to demo kateway webhook feature. |
|
Package gateway implements the full PubSub gateway features.
|
Package gateway implements the full PubSub gateway features. |
|
Package hh provides a hinted handoff service for Pub.
|
Package hh provides a hinted handoff service for Pub. |
|
disk
Package disk implements a disk-backend hinted handoff which uses raft for replication.
|
Package disk implements a disk-backend hinted handoff which uses raft for replication. |
|
kafka
Package kafka implements a kafka-backend hinted handoff.
|
Package kafka implements a kafka-backend hinted handoff. |
|
Deprecated pkg Package inflight provides storage for manipulating inflight message offsets.
|
Deprecated pkg Package inflight provides storage for manipulating inflight message offsets. |
|
Package job implements the schedulable message(job) underlying storage.
|
Package job implements the schedulable message(job) underlying storage. |
|
mysql
Package mysql implements a job store with mysql as backend.
|
Package mysql implements a job store with mysql as backend. |
|
Package manager bridges data with pubsub manager.
|
Package manager bridges data with pubsub manager. |
|
Package meta manages the global topology information.
|
Package meta manages the global topology information. |