[improve][pip] PIP-434: Expose Netty channel configuration WRITE_BUFFER_WATER_MARK to pulsar conf and pause receive requests when channel is unwritable

pip/pip-434.md

@@ -0,0 +1,95 @@
+# PIP-434: Expose Netty channel configuration WRITE_BUFFER_WATER_MARK to pulsar conf and pause receive requests when channel is unwritable
+
+# Background knowledge & Motivation
+
+As we discussed along the discussion: https://lists.apache.org/thread/6jfs02ovt13mnhn441txqy5m6knw6rr8
+
+> Problem Statement:
+> We've encountered a critical issue in our Apache Pulsar clusters where brokers experience Out-Of-Memory (OOM) errors and continuous restarts under specific load patterns. This occurs when Netty channel write buffers become full, leading to a buildup of unacknowledged responses in the broker's memory.
+
+> Background:
+> Our clusters are configured with numerous namespaces, each containing approximately 8,000 to 10,000 topics. Our consumer applications are quite large, with each consumer using a regular expression (regex) pattern to subscribe to all topics within a namespace.
+
+> The problem manifests particularly during consumer application restarts. When a consumer restarts, it issues a getTopicsOfNamespace request. Due to the sheer number of topics, the response size is extremely large. This massive response overwhelms the socket output buffer, causing it to fill up rapidly. Consequently, the broker's responses get backlogged in memory, eventually leading to the broker's OOM and subsequent restart loop.
+
+> Solution we got:
+> - Expose Netty channel configuration WRITE_BUFFER_WATER_MARK to pulsar conf
+> - Stops receive requests continuously once the Netty channel is unwritable, users can use the new config to control the threshold that limits the max bytes that are pending write.
+
+# Goals
+
+## In Scope
+- Expose Netty channel configuration WRITE_BUFFER_WATER_MARK to pulsar conf
+- Stops receive requests continuously once the Netty channel is unwritable, users can use the new config to control the threshold that limits the max bytes that are pending write.
+
+## Out of Scope
+
+- This proposal is not in order to add a broker level memory limitation, it only focuses on addressing the OOM caused by the accumulation of a large number of responses in memory due to the channel granularity being unwritable. 
+
+# Detailed Design
+
+### Configuration
+
+```shell
+# It relates to configuration "WriteBufferHighWaterMark" of Netty Channel Config. If the number of bytes queued in the write buffer exceeds this value, channel writable state will start to return "false".
+pulsarChannelWriteBufferHighWaterMark=64k
+# It relates to configuration "WriteBufferLowWaterMark" of Netty Channel Config. If the number of bytes queued in the write buffer is smaller than this value, channel writable state will start to return "true".
+pulsarChannelWriteBufferLowWaterMark=32k
+# Once the writer buffer is full, the channel stops dealing with new requests until it changes to writable
+pulsarChannelPauseReceivingRequestsIfUnwritable=false
+After the connection is recovered from an unreadable state, the channel will be rate-limited for a period of time to avoid overwhelming due to the backlog of requests. This parameter defines how many" requests should be allowed in the rate limiting period.
+pulsarChannelRateLimitingRateAfterResumeFromUnreadable=1000
+After the connection is recovered from an unreadable state, the channel will be rate-limited for a period of time to avoid overwhelming due to the backlog of requests. This parameter defines how long the rate limiting should last, in seconds. Once the bytes that are waiting to be sent out reach the pulsarChannelWriteBufferHighWaterMark\"， the timer will be reset.
+pulsarChannelRateLimitingSecondsAfterResumeFromUnreadable=5
+```
+
+### How it works
+With the settings `pulsarChannelPauseReceivingRequestsIfUnwritable=false`, the behaviour is exactly the same as the previous.
+
+After setting `pulsarChannelPauseReceivingRequestsIfUnwritable` to `true`, the channel state will be changed as follows.
+- Netty sets `channel.writable` to `false` when there is too much data that is waiting to be sent out(the size of the data cached in `ChannelOutboundBuffer` is larger than `{pulsarChannelWriteBufferHighWaterMark}`)
+  - Netty will trigger an event `channelWritabilityChanged`
+- Stops receiving requests that come into the channel, which relies on the attribute `autoRead` of the `Netty channel`.
+- Netty sets `channel.writable` to `true` once the size of the data that is waiting to be sent out is less than `{pulsarChannelWriteBufferLowWaterMark}`
+- Starts to receive requests(sets `channel.autoRead` to `true`).
+  - Note: relies on `ServerCnxThrottleTracker`, which will track the "throttle count". When a throttling condition is present, the throttle count is increased and when it's no more present, the count is decreased. The autoread should be switched to false when the counter value goes from 0 to 1, and only when it goes back from 1 to 0 should it be set to true again. The autoread flag is no longer controlled directly from the rate limiters. Rate limiters are only responsible for their part, and it's ServerCnxThrottleTracker that decides when the autoread flag is toggled.  See more details [pip-322](https://github.com/apache/pulsar/blob/master/pip/pip-322.md)
+  - To avoid handling a huge request in the backlog instantly, Pulsar will start a timed rate-limiter, which limits the rate of handling the request backlog("pulsarChannelRateLimitingRateAfterResumeFromUnreadable" requests per second).
+  - After "{pulsarChannelRateLimitingSecondsAfterResumeFromUnreadable}" seconds, the rate-limiter will be removed automatically. Once the bytes that are waiting to be sent out reach the pulsarChannelWriteBufferHighWaterMark\"， the timer will be reset.
+
+### CLI
+
+### Metrics
+| Name                                                 | Description                                                                                 | Attributes   | Units|
+|------------------------------------------------------|---------------------------------------------------------------------------------------------|--------------| --- |
+| `pulsar_server_channel_write_buf_memory_used_bytes` | Counter. The memory amount that is occupied by netty write buffers                      | cluster | - |
+
+
+# Monitoring
+
+
+# Security Considerations
+Nothing.
+
+# Backward & Forward Compatibility
+
+## Upgrade
+Nothing.
+
+## Downgrade / Rollback
+Nothing.
+
+## Pulsar Geo-Replication Upgrade & Downgrade/Rollback Considerations
+Nothing.
+
+# Alternatives
+Nothing.
+
+# General Notes
+
+# Links
+
+<!--
+Updated afterwards
+-->
+* Mailing List discussion thread: https://lists.apache.org/thread/hnbm9q3yvyf2wcbdggxmjzhr9boorqkn
+* Mailing List voting thread: https://lists.apache.org/thread/vpvtf4jnbbrhsy9y5fg00mpz9qhb0cp5