US20190238637A1

US20190238637A1 - Data replication in scalable messaging system

Info

Publication number: US20190238637A1
Application number: US16/382,090
Authority: US
Inventors: Younes Hafri; Lev Walkin; Fredrik Erik Linder
Original assignee: Satori Worldwide LLC
Current assignee: Satori Worldwide LLC
Priority date: 2016-08-31
Filing date: 2019-04-11
Publication date: 2019-08-01
Also published as: JP2019532399A; WO2018044730A1; EP3507946A1; CN109644155A; AU2017321255A1; US10305981B2; US20180063241A1

Abstract

A plurality of channels are maintained, wherein each channel comprises a plurality of messages stored in one or more buffers, wherein a first node stores a first copy of a first buffer for a first channel of the plurality of channels, wherein each of a plurality of secondary nodes stores a respective second copy of the first buffer, and wherein at least one subscriber is subscribed to messages of the first channel. A failure of the first node is detected. One of the plurality of secondary nodes are identified as a second node to access a second copy of the first buffer. The at least one subscriber is redirected from the first node to the second node and messages stored in the second copy on the second node are retrieved for the at least one subscriber.

Description

CROSS REFERENCE TO RELATED APPLICATION

This application is a continuation of U.S. patent application Ser. No. 15/252,989, filed on Aug. 31, 2016, the entire contents of which are incorporated by reference herein.

BACKGROUND

This specification relates to a data communication system and, in particular, to a system that implements data replication in a messaging system.
The publish-subscribe pattern (or “PubSub”) is a data communication messaging arrangement implemented by software systems where so-called publishers publish messages to topics and so-called subscribers receive the messages pertaining to particular topics to which they are subscribed. There can be one or more publishers per topic and publishers generally have no knowledge of which subscribers, if any, will receive the published messages. Some PubSub systems do not cache messages or have small caches meaning that subscribers may not receive messages that were published before the time of subscription to a particular topic. PubSub systems can be susceptible to performance instability during surges of message publications or as the number of subscribers to a particular topic increases.

SUMMARY

In general, one aspect of the subject matter described in this specification can be embodied in methods that include the actions of maintaining a plurality of channels, wherein each channel comprises a plurality of messages stored in one or more buffers, wherein a first node stores a first copy of a first buffer for a first channel of the plurality of channels, wherein each of a plurality of secondary nodes stores a respective second copy of the first buffer, and wherein at least one subscriber is subscribed to messages of the first channel. The method includes detecting, by one or more computer processors, a failure of the first node and identifying, by the one or more computer processors, one of the plurality of secondary nodes as a second node to access a second copy of the first buffer. The method further includes redirecting, by the one or more computer processors, the at least one subscriber from the first node to the second node; and retrieving messages stored in the second copy on the second node for the at least one subscriber. Other embodiments of this aspect include corresponding systems, apparatus, and computer programs.
The details of one or more embodiments of the subject matter described in this specification are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of the subject matter will become apparent from the description, the drawings, and the claims.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1A illustrates an example system that supports the PubSub communication pattern.

FIG. 1B illustrates functional layers of software on an example client device.

FIG. 2 is a diagram of an example messaging system.

FIG. 3A is a data flow diagram of an example method for writing data to a streamlet.

FIG. 3B is a data flow diagram of an example method for reading data from a streamlet.

FIG. 4A is a data flow diagram of an example method for publishing messages to a channel of a messaging system.

FIG. 4B is a data flow diagram of an example method for subscribing to a channel of a messaging system.

FIG. 4C is an example data structure for storing messages of a channel of a messaging system.

FIG. 5A is a data flow diagram of an example method for publishing and replicating messages of a messaging system.

FIG. 5B is a data flow diagram of an example method for retrieving stored messages in a messaging system.

FIG. 5C is a data flow diagram of an example method for repairing a chain of copies of data in a messaging system.

FIG. 5D is a data flow diagram of an example method for repairing a chain of copies of data in a messaging system.

FIG. 6 is a flowchart of an example method for publishing and subscribing messages for publishers and subscribers of a messaging system.

DETAILED DESCRIPTION

FIG. 1A illustrates an example system 100 that supports the PubSub communication pattern. Publishers (e.g., Publishers 1-N) can publish messages to named channels (e.g., Channels 1-N) by way of the system 100 (also referred to as “messaging system” hereafter). A message can include any type of information including one or more of the following: text, image content, sound content, multimedia content, video content, binary data, and so on. Other types of message data are possible. Subscribers (e.g., Subscribers 1-N) can subscribe to a named channel using the system 100 and start receiving messages which occur after the subscription request or from a given position (e.g., a message number or time offset). A client can be both a publisher and a subscriber.
Depending on the configuration, a PubSub system can be categorized as follows:

- One to One (1:1). In this configuration there is one publisher and one subscriber per channel. An example use case is private messaging.
- One to Many (1:N). In this configuration there is one publisher and multiple subscribers per channel. Example use cases are broadcasting messages (e.g., stock prices).
- Many to Many (M:N). In this configuration there are multiple publishers publishing to a single channel. The messages are then delivered to multiple subscribers. Example use cases are map applications.

There is no separate operation needed to create a named channel. A channel is created implicitly when the channel is subscribed to or when a message is published to the channel. In some implementations, channel names can be qualified by a name space. A name space includes one or more channel names. Different name spaces can have the same channel names without causing ambiguity. The name space name can be a prefix of a channel name where the name space and channel name are separated by a dot or other suitable separator. In some implementations, name spaces can be used when specifying channel authorization settings. For instance, the system 100 may have app1.foo and app1.system.notifications channels where “app1” is the name of the name space. The system can allow clients to subscribe and publish to the app1.foo channel. However, clients can subscribe, but not publish, to the app1.system.notifications channel.
FIG. 1B illustrates functional layers of software on an example client device. A client device (e.g., client 102) is a data processing apparatus such as, for example, a personal computer, a laptop computer, a tablet computer, a smart phone, a smart watch, or a server computer. Other types of client devices are possible. The application layer 104 includes the end-user application(s) that will integrate with the system 100. The messaging layer 106 is a programmatic interface for the application layer 104 to utilize services of the system 100 such as channel subscription, message publication, message retrieval, user authentication, and user authorization. In some implementations, the messages passed to and from the messaging layer 106 are encoded as JavaScript Object Notation (JSON) objects. Other message encoding schemes are possible.
The operating system layer 108 includes the operating system software on the client 102. In various implementations, messages can be sent and received to/from the system 100 using persistent or non-persistent connections. Persistent connections can be created using, for example, network sockets. A transport protocol such as TCP/IP layer 112 implements the Transport Control Protocol/Internet Protocol communication with the system 100 that can be used by the messaging layer 106 to send messages over connections to the system 100. Other communication protocols are possible including, for example, User Datagram Protocol (UDP). In further implementations, an optional Transport Layer Security (TLS) layer 110 can be employed to ensure the confidentiality of the messages.
FIG. 2 is a diagram of an example messaging system 100. The messaging system 100 provides functionality for implementing PubSub communication patterns. The messaging system 100 includes software components and storage that can be deployed at one or more data centers 122 in one or more geographic locations, for example. The messaging system 100 includes multiplexer (MX) nodes 202, 204 and 206, queue (Q) nodes 208, 210 and 212, one or more channel manager nodes (e.g., channel managers 214, 215), and optionally one or more cache (C) nodes 220 and 222. Each node can execute in a virtual machine or on a physical machine (e.g., a data processing apparatus). Each MX node serves as a termination point for one or more publisher and/or subscriber connections through the external network 216. The internal communication among MX nodes, Q nodes, C nodes, and the channel managers, is conducted over an internal network 218. For example, MX node 204 can be the terminus of a subscriber connection from client 102. Each Q node buffers channel data for consumption by the MX nodes. An ordered sequence of messages published to a channel is a logical channel stream. For example, if three clients publish messages to a given channel, the combined messages published by the clients include a channel stream. Messages can be ordered in a channel stream. For example, the messages may be ordered by time of publication by the client, by time of receipt by an MX node, or by time of receipt by a Q node. Other ways for ordering messages in a channel stream are possible. In the case where more than one message would be assigned to the same position in the order, one of the messages can be chosen (e.g., randomly) to have a later sequence in the order. Each channel manager node is responsible for managing Q node load by splitting channel streams into streamlets, as will be discussed in further detail below. The optional C nodes provide caching and load removal from the Q nodes.
In the example messaging system 100, one or more client devices (publishers and/or subscribers) establish respective persistent connections (e.g., TCP connections) to an MX node (e.g., MX nodes 202, 204 and/or 206). The MX node serves as a termination point for these connections. For instance, external messages (e.g., between respective client devices and the MX node) carried by these connections can be encoded based on an external protocol (e.g., JSON). The MX node terminates the external protocol and translates the external messages to internal communication, and vice versa. The MX nodes 202, 204 and 206 publish and subscribe to streamlets on behalf of clients. In this way, an MX node can multiplex and merge requests of client devices subscribing for or publishing to the same channel, thus representing multiple client devices as one, instead of one by one.
In the example messaging system 100, a Q node (e.g., Q nodes 208, 210 and/or 212) can store one or more streamlets of one or more channel streams. A streamlet is a data buffer for a portion of a channel stream. A streamlet will close to writing when its storage is full. A streamlet will close to reading and writing and be de-allocated when its time-to-live (TTL) has expired. For example, a streamlet can have a maximum size of 1 MB and a TTL of three minutes. Different channels can have streamlets limited by different sizes and/or by different TTLs. For example, streamlets in one channel can exist for up to three minutes, while streamlets in another channel can exist for up to 10 minutes. In various implementations, a streamlet corresponds to a computing process running on a Q node. The computing process can be terminated after the streamlet's TTL has expired, thus freeing up computing resources (for the streamlet) back to the Q node, for example.
When receiving a publish request from client 102, an MX node (e.g., MX node 204) makes a request to a channel manager (e.g., channel manager 214) to grant access to a streamlet to write the message being published. However, if the MX node has already been granted write access to a streamlet for the channel (and the channel has not been closed to writing), the MX node can write the message to that streamlet without having to request a grant to access the streamlet. Once a message is written to a streamlet for a channel, the message can be read by MX nodes and provided to subscribers of that channel.
Similarly, when receiving a channel subscription request from a client device, an MX node makes a request to a channel manager to grant access to a streamlet for the channel from which messages are read. If the MX node has already been granted read access to a streamlet for the channel (and the channel's TTL has not been closed to reading) the MX node can read messages from the streamlet without having to request a grant to access the streamlet. The read messages can then be forwarded to client devices that have subscribed to the channel. In various implementations, messages read from streamlets are cached by MX nodes so that MX nodes can reduce the number of times messages are read from the streamlets.
In implementations, an MX node can request a grant from the channel manager that allows the MX node to store a block of data into a streamlet on a particular Q node that stores streamlets of a particular channel. Example streamlet grant request and grant data structures are as follows:


	StreamletGrantRequest = {

	“channel”: string( )
	“mode”: “read” \| “write”
	“position”: 0

	}
	StreamletGrantResponse = {

	“streamlet-id”: “abcdef82734987”,
	“limit-size”: 2000000, # 2 megabytes max
	“limit-msgs”: 5000, # 5 thousand messages max
	“limit-life”: 4000, # the grant is valid for 4 seconds
	“q-node”: string( )
	“position”: 0

	}

The StreamletGrantRequest data structure stores the name of the stream channel and a mode indicating whether the MX node intends on reading from or writing to the streamlet. The MX node sends the StreamletGrantRequest to a channel manager node. The channel manager node, in response, sends the MX node a StreamletGrantResponse data structure. The StreamletGrantResponse contains an identifier of the streamlet (streamlet-id), the maximum size of the streamlet (limit-size), the maximum number of messages that the streamlet can store (limit-msgs), the TTL (limit-life), and an identifier of a Q node (q-node) on which the streamlet resides. The StreamletGrantRequest and StreamletGrantResponse can also have a position field that points to a position in a streamlet (or a position in a channel) for reading from the streamlet.
A grant becomes invalid once the streamlet has closed. For example, a streamlet is closed to reading and writing once the streamlet's TTL has expired, and a streamlet is closed to writing when the streamlet's storage is full. When a grant becomes invalid, the MX node can request a new grant from the channel manager to read from or write to a streamlet. The new grant will reference a different streamlet and will refer to the same or a different Q node depending on where the new streamlet resides.
FIG. 3A is a data flow diagram of an example method 300 for writing data to a streamlet in various embodiments. In FIG. 3A, when an MX node's (e.g., MX node 202) request to write to a streamlet is granted by a channel manager (e.g., channel manager 214), the MX node 202 establishes a Transmission Control Protocol (TCP) connection with the Q node (e.g., Q node 208) identified in the grant response received from the channel manager (302). A streamlet can be written concurrently by multiple write grants (e.g., for messages published by multiple publishers). Other types of connection protocols between the MX node 202 and the Q node 208 are possible.
The MX node 202 sends a prepare-publish message with an identifier of a streamlet that the MX node 202 wants to write to the Q node 208 (304). The streamlet identifier and Q node identifier can be provided by the channel manager in the write grant as described earlier. The Q node 202 provides the message to a handler 301 (e.g., a computing process running on the Q node 208) for the identified streamlet (306). The handler 301 can send an acknowledgment to the MX node 202 (308). After receiving the acknowledgement, the MX node 202 starts writing (publishing) messages (e.g., 310, 312, 314, and 318) to the handler 301, which stores the received data in the identified streamlet. The handler 301 can also send acknowledgements (316, 320) to the MX node 202 for the received data. In some implementations, acknowledgements can be piggy-backed or cumulative. For example, the handler 301 can send an acknowledgement to the MX node 202 for every predetermined amount of data received (e.g., for every 100 messages received) or for every predetermined time period (e.g., for every one millisecond). Other acknowledgement scheduling algorithms, such as Nagle's algorithm, can be used.
If the streamlet can no longer accept published data (e.g., the streamlet is full), the handler 301 sends a Negative-Acknowledgement (NAK) message (330) indicating a problem, following by an EOF (end-of-file) message (332). In this way, the handler 301 closes the association with the MX node 202 for the publish grant. The MX node 202 can request a write grant for another streamlet from a channel manager if the MX node 202 has additional messages to store.
FIG. 3B is a data flow diagram of an example method 350 for reading data from a streamlet in various embodiments. In FIG. 3B, an MX node (e.g., MX node 204) sends a request to a channel manager (e.g., channel manager 214) for reading a particular channel starting from a particular message or time offset in the channel. The channel manager returns a read grant to the MX node 204 including an identifier of a streamlet containing the particular message, a position in the streamlet corresponding to the particular message, and an identifier of a Q node (e.g., Q node 208) containing the particular streamlet. The MX node 204 then establishes a TCP connection with the Q node 208 (352). Other types of connection protocols between the MX node 204 and the Q node 208 are possible.
The MX node 204 then sends a subscribe message (354) to the Q node 208 with the identifier of the streamlet in the Q node 208 and the position in the streamlet from which the MX node 204 wants to read (356). The Q node 208 provides the subscribe message to a handler 351 of the streamlet (356). The handler 351 can send an acknowledgement to the MX node 204 (358). The handler 351 sends messages (360, 364, 366), starting at the position in the streamlet, to the MX node 204. In some implementations, the handler 351 can send all of the messages in the streamlet to the MX node 204. After sending the last message in a particular streamlet, the handler 351 can send a notification of the last message to the MX node 204. The MX node 204 can send to the channel manager another request for another streamlet containing a next message in the particular channel.
If the particular streamlet is closed (e.g., after its TTL has expired), the handler 351 can send an unsubscribe message (390), followed by an EOF message (392), to close the association with the MX node 204 for the read grant. The MX node 204 can close the association with the handler 351 when the MX node 204 moves to another streamlet for messages in the particular channel (e.g., as instructed by the channel manager). The MX node 204 can also close the association with the handler 351 if the MX node 204 receives an unsubscribe message from a corresponding client device.
In various implementations, a streamlet can be written into and read from at the same time instance. For example, there can be a valid read grant and a valid write grant at the same time instance. In various implementations, a streamlet can be read concurrently by multiple read grants (e.g., for channels subscribed to by multiple publisher clients). The handler of the streamlet can order messages from concurrent write grants based on, for example, time-of-arrival, and store the messages based on the order. In this way, messages published to a channel from multiple publisher clients can be serialized and stored in a streamlet of the channel.
In the messaging system 100, one or more C nodes (e.g., C node 220) can offload data transfers from one or more Q nodes. For example, if there are multiple MX nodes requesting streamlets from Q nodes for a particular channel, the streamlets can be offloaded and cached in one or more C nodes. The MX nodes (e.g., as instructed by read grants from a channel manager) can read the streamlets from the C nodes instead.
As described above, messages for a channel in the messaging system 100 are ordered in a channel stream. A channel manager (e.g., channel manager 214) splits the channel stream into fixed-sized streamlets that each reside on a respective Q node. In this way, storing a channel stream can be shared among many Q nodes; each Q node stores a portion (one or more streamlets) of the channel stream. More particularly, a streamlet can be stored in, for example, registers and/or dynamic memory elements associated with a computing process on a Q node, thus avoiding the need to access persistent, slower storage devices such as hard disks. This results in faster message access. The channel manager can also balance loads among Q nodes in the messaging system 100 by monitoring respective workloads of the Q nodes and allocating streamlets in a way that avoids overloading any one Q node.
In various implementations, a channel manager maintains a list identifying each active streamlet, the respective Q node on which the streamlet resides, an identification of the position of the first message in the streamlet, and whether the streamlet is closed for writing. In some implementations, Q nodes notify the channel manager and any MX nodes that are publishing to a streamlet that the streamlet is closed due to being full or when the streamlet's TTL has expired. When a streamlet is closed, the streamlet remains on the channel manager's list of active streamlets until the streamlet's TTL has expired so that MX nodes can continue to retrieve messages from the streamlet.
When an MX node requests a write grant for a given channel and there is not a streamlet for the channel that can be written to, the channel manager allocates a new streamlet on one of the Q nodes and returns the identity of the streamlet and the Q node in the StreamletGrantResponse to the MX node. Otherwise, the channel manager returns the identity of the currently open for writing streamlet and corresponding Q node in the StreamletGrantResponse to the MX node. MX nodes can publish messages to the streamlet until the streamlet is full or the streamlet's TTL has expired, after which a new streamlet can be allocated by the channel manager.
When an MX node requests a read grant for a given channel and there is not a streamlet for the channel that can be read from, the channel manager allocates a new streamlet on one of the Q nodes and returns the identity of the streamlet and the Q node in the StreamletGrantResponse to the MX node. Otherwise, the channel manager returns the identity of the streamlet and Q node that contains the position from which the MX node wishes to read to the MX node. The Q node can then begin sending messages to the MX node from the streamlet beginning at the specified position until there are no more messages in the streamlet to send. When a new message is published to a streamlet, MX nodes that have subscribed to that streamlet will receive the new message. If a streamlet's TTL has expired, the handler process 351 sends an EOF message (392) to any MX nodes that are subscribed to the streamlet.
As described earlier in reference to FIG. 2, the messaging system 100 can include multiple channel managers (e.g., channel managers 214, 215). Multiple channel managers provide resiliency and prevent single point of failure. For instance, one channel manager can replicate lists of streamlets and current grants it maintains to another “slave” channel manager. As for another example, multiple channel managers can coordinate operations between them using distributed consensus protocols, such as, for example, Paxos or Raft protocols.
FIG. 4A is a data flow diagram of an example method 400 for publishing messages to a channel of a messaging system. In FIG. 4A, publishers (e.g., publishers 402, 404, 406) publish messages to the messaging system 200 described earlier in reference to FIG. 2. For instance, publishers 402 respectively establish connections 411 and send publish requests to the MX node 202. Publishers 404 respectively establish connections 413 and send publish requests to the MX node 206. Publishers 406 respectively establish connections 415 and send publish requests to the MX node 204. Here, the MX nodes can communicate (417) with a channel manager (e.g., channel manager 214) and one or more Q nodes (e.g., Q nodes 212 and 208) in the messaging system 100 via the internal network 218.
By way of illustration, each publish request (e.g., in JSON key/value pairs) from a publisher to an MX node includes a channel name and a message. The MX node (e.g., MX node 202) can assign the message in the publish request to a distinct channel in the messaging system 100 based on the channel name (e.g., “foo”) of the publish request. The MX node can confirm the assigned channel with the channel manager 214. If the channel (specified in the subscribe request) does not yet exist in the messaging system 100, the channel manager can create and maintain a new channel in the messaging system 100. For instance, the channel manager can maintain a new channel by maintaining a list identifying each active streamlet of the channel's stream, the respective Q node on which the streamlet resides, and identification of the positions of the first and last messages in the streamlet as described earlier.
For messages of a particular channel, the MX node can store the messages in one or more buffers or streamlets in the messaging system 100. For instance, the MX node 202 receives, from the publishers 402, requests to publish messages M11, M12, M13, and M14 to a channel foo. The MX node 206 receives, from the publishers 404, requests to publish messages M78 and M79 to the channel foo. The MX node 204 receives, from the publishers 406, requests to publish messages M26, M27, M28, M29, M30, and M31 to the channel foo.
The MX nodes can identify one or more streamlets for storing messages for the channel foo. As described earlier, each MX node can request a write grant from the channel manager 214 that allows the MX node to store the messages in a streamlet of the channel foo. For instance, the MX node 202 receives a grant from the channel manager 214 to write messages M11, M12, M13, and M14 to a streamlet 4101 on the Q node 212. The MX node 206 receives a grant from the channel manager 214 to write messages M78 and M79 to the streamlet 4101. Here, the streamlet 4101 is the last streamlet of a sequence of streamlets of the channel stream 430 storing messages of the channel foo. The streamlet 4101 has messages 421 of the channel foo that were previously stored in the streamlet 4101, but is still open (e.g., the streamlet 4101 still has space for storing more messages and the streamlet's TTL has not expired.)
The MX node 202 can arrange the messages for the channel foo based on the respective time that each of the messages 422 was received by the MX node 202 (e.g., M11, M13, M14, M12) and store the received messages as arranged in the streamlet 4101. That is, the MX node 202 receives M11 first, followed by M13, M14, and M12. Similarly, the MX node 206 can arrange the messages for the channel foo based on their respective time that each of the messages 423 was received by the MX node 206 (e.g., M78, M79) and store the received messages 423 as arranged in the streamlet 4101. Other arrangements or ordering of the messages for the channel are possible.
The MX node 202 (or MX node 206) can store the received messages using the method for writing data to a streamlet described earlier in reference to FIG. 3A, for example. In various implementations, the MX node 202 (or MX node 206) can buffer (e.g., in a local data buffer) the received messages for the channel foo and store the received messages in a streamlet for the channel foo (e.g., streamlet 4101) when the buffered messages reach a predetermined number or size (e.g., 100 messages) or when a predetermined time (e.g., 50 milliseconds) has elapsed. For example, the MX node 202 can store in the streamlet 100 messages at a time or in 50 millisecond increments. Other acknowledgement scheduling algorithms, such as Nagle's algorithm, can be used.
In various implementations, the Q node 212 (e.g., a handler) stores the messages of the channel foo in the streamlet 4101 in the order as arranged by the MX node 202 and MX node 206. The Q node 212 stores the messages of the channel foo in the streamlet 4101 in the order the Q node 212 receives the messages. For instance, assume that the Q node 212 receives message M78 (from the MX node 206) first, followed by messages M11 and M13 (from the MX node 202), M79 (from the MX node 206), and M14 and M12 (from the MX node 202). The Q node 212 stores in the streamlet 4101 the messages in the order as received (e.g., M78, M11, M13, M79, M14, and M12) immediately after the messages 421 that are already stored in the streamlet 4101. In this way, messages published to the channel foo from multiple publishers (e.g., MX nodes 402, 404) can be serialized in a particular order and stored in the streamlet 4101 of the channel foo. Different subscribers that subscribe to the channel foo will receive messages of the channel foo in the same particular order, as will be described in more detail in reference to FIG. 4B.
In the example of FIG. 4A, at a time instance after the message M12 was stored in the streamlet 4101, the MX node 204 requests a grant from the channel manager 214 to write to the channel foo. The channel manager 214 provides the MX node 204 a grant to write messages to the streamlet 4101, as the streamlet 4101 is still open for writing. The MX node 204 arranges the messages for the channel foo based on the respective time that each message 424 was received by the MX node 204 (e.g., M26, M27, M31, M29, M30, M28) and stores the messages as arranged for the channel foo.
By way of illustration, assume that the message M26 is stored to the last available position of the streamlet 4101. As the streamlet 4101 is now full, the Q node 212 sends to the MX node 204 a NAK message, following by an EOF message, to close the association with the MX node 204 for the write grant, as described earlier in reference to FIG. 3A. The MX node 204 then requests another write grant from the channel manager 214 for additional messages (e.g., M27, M31, and so on) for the channel foo.
The channel manager 214 can monitor available Q nodes in the messaging system 100 for the Q nodes respective workloads (e.g., how many streamlets are residing in each Q node). The channel manager 214 can allocate a streamlet for the write request from the MX node 204 such that overloading (e.g., too many streamlets or too many read or write grants) can be avoided for any given Q node. For example, the channel manager 214 can identify a least loaded Q node in the messaging system 100 and allocate a new streamlet on the least loaded Q node for write requests from the MX node 204. In the example of FIG. 4A, the channel manager 214 allocates a new streamlet 4102 on the Q node 208 and provides a write grant to the MX node 204 to write messages for the channel foo to the streamlet 4102. As shown in FIG. 4A, the Q node 208 stores in the streamlet 4102 the messages from the MX node 204 in an order as arranged by the MX node 204: M27, M31, M29, M30, and M28 (assuming that there is no other concurrent write grants for the streamlet 4102 at the moment).
When the channel manager 214 allocates a new streamlet (e.g., streamlet 4102) for a request for a grant from an MX node (e.g., MX node 204) to write to a channel (e.g., foo), the channel manager 214 assigns to the streamlet its TTL, which will expire after TTLs of other streamlets that are already in the channel's stream. For instance, the channel manager 214 can assign to each streamlet of the channel foo's channel stream a TTL of 3 minutes when allocating the streamlet. That is, each streamlet will expire 3 minutes after it is allocated (created) by the channel manager 214. Since a new streamlet is allocated after a previous streamlet is closed (e.g., filled entirely or expired), in this way, the channel foo's channel stream includes streamlets that each expire sequentially after the previous streamlet expires. For example, as shown in an example channel stream 430 of the channel foo in FIG. 4A, streamlet 4098 and streamlets before 4098 (e.g., streamlet 4097) have expired (as indicated by the dotted-lined gray-out boxes). Messages stored in these expired streamlets are not available for reading for subscribers of the channel foo. Streamlets 4099, 4100, 4101, and 4102 are still active (not expired). The streamlets 4099, 4100, and 4101 are closed for writing, but still are available for reading. The streamlet 4102 is available for reading and writing, at the moment when the message M28 was stored in the streamlet 4102. At a later time, the streamlet 4099 will expire, following by the streamlets 4100, 4101, and so on.
FIG. 4B is a data flow diagram of an example method 450 for subscribing to a channel of a messaging system. In FIG. 4B, a subscriber 480 establishes a connection 462 with an MX node 461 of the messaging system 100. Subscriber 482 establishes a connection 463 with the MX node 461. Subscriber 485 establishes a connection 467 with an MX node 468 of the messaging system 100. Here, the MX nodes 461 and 468 can respectively communicate 464 with the channel manager 214 and one or more Q nodes in the messaging system 100 via the internal network 218.
A subscriber (e.g., subscriber 480) can subscribe to the channel foo of the messaging system 100 by establishing a connection 462 and sending a request for subscribing to messages of the channel foo to an MX node (e.g., MX node 461). The request (e.g., in JSON key/value pairs) can include a channel name, such as, for example, “foo.” When receiving the subscribe request, the MX node 461 can send a request to the channel manager 214 for a read grant for a streamlet in the channel foo's channel stream.
By way of illustration, assume that at the current moment the channel foo's channel stream 431 includes active streamlets 4102, 4103, and 4104, as shown in FIG. 4B. The streamlets 4102 and 4103 each are full. The streamlet 4104 stores messages of the channel foo, including the last message stored at a position 47731. Streamlets 4101 and streamlets before 4101 are invalid, as their respective TTLs have expired. Note that the messages M78, M11, M13, M79, M14, M12, and M26 stored in the streamlet 4101, described earlier in reference to FIG. 4A, are no longer available for subscribers of the channel foo, since the streamlet 4101 is no longer valid, as the TTL of streamlet 4101 has expired. As described earlier, each streamlet in the channel foo's channel stream has a TTL of 3 minutes, thus only messages (as stored in streamlets of the channel foo) that are published to the channel foo (i.e., stored into the channel's streamlets) no earlier than 3 minutes from the current time can be available for subscribers of the channel foo.
The MX node 461 can request a read grant for all available messages in the channel foo, for example, when the subscriber 480 is a new subscriber to the channel foo. Based on the request, the channel manager 214 provides the MX node 461 a read grant to the streamlet 4102 (on the Q node 208) that is the earliest streamlet in the active streamlets of the channel foo (e.g., the first in the sequence of the active streamlets). The MX node 461 can retrieve messages in the streamlet 4102 from the Q node 208, using the method for reading data from a streamlet described earlier in reference to FIG. 3B, for example. Note that the messages retrieved from the streamlet 4102 maintain the same order as stored in the streamlet 4102. However, other arrangements or ordering of the messages in the streamlet are possible. In various implementations, when providing messages stored in the streamlet 4102 to the MX node 461, the Q node 208 can buffer (e.g., in a local data buffer) the messages and send the messages to the MX node 461 when the buffer messages reach a predetermined number or size (e.g., 200 messages) or a predetermined time (e.g., 50 milliseconds) has elapsed. For instance, the Q node 208 can send the channel foo's messages (from the streamlet 4102) to the MX node 461 200 messages at a time or in 50 millisecond increments. Other acknowledgement scheduling algorithms, such as Nagle's algorithm, can be used.
After receiving the last message in the streamlet 4102, the MX node 461 can send an acknowledgement to the Q node 208, and send to the channel manager 214 another request (e.g., for a read grant) for the next streamlet in the channel stream of the channel foo. Based on the request, the channel manager 214 provides the MX node 461 a read grant to the streamlet 4103 (on Q node 472) that logically follows the streamlet 4102 in the sequence of active streamlets of the channel foo. The MX node 461 can retrieve messages stored in the streamlet 4103 using the method for reading data from a streamlet described earlier in reference to FIG. 3B, until MX node 461 retrieves the last message stored in the streamlet 4103. The MX node 461 can send to the channel manager 214 yet another request for a read grant for messages in the next streamlet 4104 (on Q node 474). After receiving the read grant, the MX node 461 retrieves message of the channel foo stored in the streamlet 4104, until the last message at the position 47731 is retrieved by MX node 461. Similarly, the MX node 468 can retrieve messages from the streamlets 4102, 4103, and 4104 (as shown with dotted arrows in FIG. 4B), and provide the messages to the subscriber 485.
The MX node 461 can send the retrieved messages of the channel foo to the subscriber 480 (via the connection 462) while receiving the messages from the Q node 208, 472, or 474. In various implementations, the MX node 461 can store the retrieved messages in a local buffer. In this way, the retrieved messages can be provided to another subscriber (e.g., subscriber 482) when the other subscriber subscribes to the channel foo and requests the channel's messages. The MX node 461 can remove messages stored in the local buffer that each has a time of publication that has exceeded a predetermined time period. For instance, the MX node 461 can remove messages stored in the local buffer with respective times of publication exceeding 3 minutes. In some implementations, the predetermined time period for keeping messages in the local buffer on MX node 461 can be the same as or similar to the time-to-live duration of a streamlet in the channel foo's channel stream, since at a given moment, messages retrieved from the channel's stream do not include the messages in streamlets having respective time-to-lives that have already expired.
The messages retrieved from the channel stream 431 and sent to the subscriber 480 (by the MX node 461) are arranged in the same order as the messages were stored in the channel stream, although other arrangements or ordering of the messages are possible. For example, messages published to the channel foo are serialized and stored in the streamlet 4102 in a particular order (e.g., M27, M31, M29, M30, and so on), then stored subsequently in the streamlet 4103 and the streamlet 4104. The MX node 461 retrieves messages from the channel stream 431 and provides the retrieved messages to the subscriber 480 in the same order as the messages are stored in the channel stream: M27, M31, M29, M30, and so on, followed by ordered messages in the streamlet 4103, and followed by ordered messages in the streamlet 4104.
Instead of retrieving all available messages in the channel stream 431, the MX node 461 can request a read grant for messages stored in the channel stream 431 starting from a message at particular position (e.g., position 47202.) For example, the position 47202 can correspond to an earlier time instance (e.g., 10 seconds before the current time) when the subscriber 480 was last subscribing to the channel foo (e.g., via a connection to the MX node 461 or another MX node of the messaging system 100). The MX node 461 can send a request to the channel manager 214 for a read grant for messages starting at the position 47202. Based on the request, the channel manager 214 provides the MX node 461 a read grant to the streamlet 4104 (on the Q node 474) and a position on the streamlet 4104 that corresponds to the channel stream position 47202. The MX node 461 can retrieve messages in the streamlet 4104 starting from the provided position, and send the retrieved messages to the subscriber 480.
As described above in reference to FIGS. 4A and 4B, messages published to the channel foo are serialized and stored in the channel's streamlets in a particular order. The channel manager 214 maintains the ordered sequence of streamlets as they are created throughout their respective time-to-lives. Messages retrieved from the streamlets by an MX node (e.g., MX node 461 and/or MX node 468) and provided to a subscriber can be, in some implementations, in the same order as the messages are stored in the ordered sequence of streamlets. In this way, messages sent to different subscribers (e.g., subscriber 480, subscriber 482, or subscriber 485) can be in the same order as the messages are stored in the streamlets, regardless which MX nodes the subscribers are connected to.
In various implementations, a streamlet stores messages in a set of blocks of messages. Each block stores a number of messages. For instance, a block can store two hundred kilobytes of messages. Each block has its own time-to-live, which can be shorter than the time-to-live of the streamlet holding the block. Once a block's TTL has expired, the block can be discarded from the streamlet holding the block, as described in more detail below in reference to FIG. 4C.
FIG. 4C is an example data structure 490 for storing messages of a channel of a messaging system. As described with the channel foo in reference to FIGS. 4A and 4B, assume that at the current moment the channel foo's channel stream 432 includes active streamlets 4104 and 4105. Streamlet 4103 and streamlets before streamlet 4103 (e.g., streamlets 4101 and 4102) are invalid, as their respective TTLs have expired. The streamlet 4104 has reached maximum capacity (e.g., as determined by a corresponding write grant) and is closed for additional message writes. The streamlet 4104 is available for message reads. The streamlet 4105 is open and is available for message writes and reads.
By way of illustration, the streamlet 4104 (e.g., a computing process running on the Q node 474 shown in FIG. 4B) currently holds two blocks of messages. Block 494 holds messages from channel positions 47301 to 47850. Block 495 holds messages from channel positions 47851 to 48000. The streamlet 4105 (e.g., a computing process running on another Q node in the messaging system 100) currently holds two blocks of messages. Block 496 holds messages from channel positions 48001 to 48200. Block 497 holds messages starting from channel position 48201, and still accepts additional messages of the channel foo.
When the streamlet 4104 was created (e.g., by a write grant), block 492 was created to store messages from channel positions 47010 to 47100. Later on, after the block 492 had reached its capacity, another block 493 was created to store messages (e.g., from channel positions 47111 to 47300.) Blocks 494 and 495 were subsequently created to store additional messages. Afterwards, the streamlet 4104 was closed for additional message writes, and the streamlet 4105 was created with additional blocks for storing additional messages of the channel foo.
In this example, the respective TTL's of blocks 492 and 493 have expired. The messages stored in these two blocks (from channel positions 47010 to 47300) are no longer available for reading by subscribers of the channel foo. The streamlet 4104 can discard these two expired blocks. For example, the streamlet 4104 can de-allocate the memory space for the blocks 492 and 493. The blocks 494 or 495 could become expired and be discarded by the streamlet 4104, before the streamlet 4104 itself becomes invalid. Alternatively, streamlet 4104 itself could become invalid before the blocks 494 or 495 become expired. For example, a streamlet can hold one or more blocks of messages, or contain no block of messages, depending on respective TTLs of the streamlet and blocks.
A streamlet, or a computing process running on a Q node in the messaging system 100, can create a block for storing messages of a channel by allocating a certain size of memory space from the Q node. The streamlet can receive, from an MX node in the messaging system 100, one message at a time and store the received message in the block. Alternatively, the MX node can assemble (e.g., buffer) a group of messages and send the group of messages to the Q node. The streamlet can allocate a block of memory space from the Q node and store the group of messages in the block. The MX node can also perform compression on the group of messages. For example, the MX node can remove a common header from each message or performing other suitable compression techniques.
As described above, a streamlet (e.g., a data buffer) residing on a Q node stores messages of a channel in the messaging system 100. To prevent failure of the Q node (a single point failure) causing messages to be lost, the messaging system 100 can replicate messages on multiple Q nodes, as described in more detail below.
FIG. 5A is a data flow diagram of an example method 500 for publishing and replicating messages of the messaging system 100. As described earlier in reference to FIG. 4A, the MX node 204 receives messages (of the channel foo) from the publishers 406. The channel manager 214 can instruct the MX node 204 (e.g., with a write grant) to store the messages in the streamlet 4102 on the Q node 208. In FIG. 5A, instead of storing the messages on a single node (e.g., Q node 208), the channel manager 214 allocates multiple Q nodes to store multiple copies of the streamlet 4102 on these Q nodes.
By way of illustration, the channel manager 214 allocates Q nodes 208, 502, 504, and 506 in the messaging system 100 to store copies of the streamlet 4102. The channel manager 214 instructs the MX node 204 to transmit the messages for the channel foo (e.g., messages M27, M31, M29, M30, and M28) to the Q node 208 (512). A computing process running on the Q node 208 stores the messages in the first copy (copy #1) of the streamlet 4102. Instead of sending an acknowledgement message to the MX node 204 after storing the messages, the Q node 208 forwards the messages to the Q node 502 (514). A computing process running on the Q node 502 stores the messages in another copy (copy #2) of the streamlet 4102. Meanwhile, the Q node 502 forwards the messages to the Q node 504 (516). A computing process running on the Q node 504 stores the messages in yet another copy (copy #3) of the streamlet 4102. The Q node 504 also forwards the message to the Q node 506 (518). A computing process running on the Q node 506 stores the messages in yet another copy (copy #4) of the streamlet 4102. The Q node 506 can send an acknowledgement message to the MX node 204, indicating that all the messages (M27, M31, M29, M30, and M28) have been stored successfully in streamlet copies #1, #2, #3 and #4.
In some implementations, after successfully storing the last copy (copy #4), the Q node 506 can send an acknowledgement to upstream Q node 504, which in turns sends an acknowledgement to upstream Q node 502, and so on, until the acknowledgement is sent to the Q node 208 storing the first copy (copy #1). The Q node 208 can send an acknowledgement message to the MX node 204, indicating that all messages have been stored successfully in the streamlet 4102 (i.e., in the copies #1, #2, #3 and #4).
In this way, four copies of the streamlet 4102 (and each message in the streamlet) are stored in four different Q nodes. Other numbers (e.g., two, three, five, or other suitable number) of copies of a streamlet are also possible. In the present illustration, the four copies form a chain of copies including a head copy in the copy #1 and a tail copy in the copy #4. When a new message is published to the streamlet 4102, the message is first stored in the head copy (copy #1) on the Q node 208. The message is then forwarded downstream to the next adjacent copy, the copy #2 on the Q node 502 for storage, then to the copy #3 on the Q node 504 for storage, until the message is stored in the tail copy the copy #4 on the Q node 506.
In addition to storing and forwarding by messages, the computing processes running on Q nodes that store copies of a streamlet can also store and forward messages by blocks of messages, as described earlier in reference to FIG. 4C. For instance, the computing process storing the copy #1 of the streamlet 4102 on Q node 208 can allocate memory and store a block of, for example, 200 kilobytes of messages (although other sizes of blocks of messages are possible), and forward the block of messages to the next adjacent copy (copy #2) of the chain for storage, and so on, until the block messages is stored in the tail copy (copy #4) on the Q node 506.
Messages of the streamlet 4102 can be retrieved and delivered to a subscriber of the channel foo from one of the copies of the streamlet 4102. FIG. 5B is a data flow diagram of an example method 550 for retrieving stored messages in the messaging system 100. For instance, the subscriber 480 can send a request for subscribing to messages of the channel to the MX node 461, as described earlier in reference to FIG. 4B. The channel manager 214 can provide a read grant to the MX node 461 for one of the copies of the streamlet 4102. The MX Node 461 can retrieve messages of the streamlet 4102 from one of the Q nodes storing a copy of the streamlet 4102, and provide the retrieved messages to the subscriber 480. For instance, the MX node 461 can retrieve messages from the copy #4 (the tail copy) stored on the Q node 506 (522). As for another example, the MX node 461 can retrieve messages from the copy #2 stored on the Q node 502 (524). In this way, the multiple copies of a streamlet (e.g., copies #1, #2, #3, and #4 of the streamlet 4102) provide replication and redundancy against failure of one or more of Q nodes 208, 502, 504 or 506 of the messaging system 100. In various implementations, the channel manager 214 can balance workloads among the Q nodes storing copies of the streamlet 4102 by directing the MX node 461 (e.g., with a read grant) to a particular Q node. For example, the channel manager 214 can direct MX node 461 to a Q node having less current read and write grants as compared to other Q nodes storing copies of the streamlet 4102.
In an example scenario, a Q node storing a particular copy in a chain of copies of a streamlet may fail. For example, a computing process on the Q node storing the particular copy may freeze. Other failure modes of a Q node are possible. An MX node can detect a failed node (e.g., from non-responsiveness of the failed node) and report the failed node to a channel manager in the messaging system 100 (e.g., channel manager 214). A peer Q node can also detect a failed Q node and report the failed node to the channel manager. For instance, an upstream Q node may detect a failed downstream Q node when the downstream Q node is non-responsive. For example, the downstream Q node may fail to acknowledge a message storage request from the upstream Q node. It is noted that failure of a Q node storing a copy of a particular streamlet of a particular channel stream does not have to be for publish or subscribe operations of the particular streamlet or of the particular channel stream. Failure stemming from operations on another streamlet or another channel stream can also alert a channel manager about failure of a Q node in the messaging system 100.
When a Q node storing a particular copy in a chain of copies of a streamlet fails, a channel manager in the messaging system 100 can repair the chain. For example, the messaging system 100 can repair the chain by removing the failed node or by inserting a new node for a new copy into the chain. FIG. 5C is a data flow diagram of an example method 570 for repairing a chain of copies of a streamlet in the messaging system 100. In FIG. 5C, for instance, after detecting that the Q node 504 fails, the channel manager 214 can repair the chain of copies by redirecting messages intended to be stored in the copy #3 of the streamlet 4102 on the Q node 502 to the copy #4 of the streamlet 4102 on the Q node 506. In this example, a message (or a block of messages) is first sent from the MX node 204 to the Q node 208 for storage in the copy #1 of the streamlet 4102 (572). The message then is forwarded to the Q node 502 for storage in the copy #2 of the streamlet 4102 (574). The message is then forwarded to the Q node 506 for storage in the copy #4 of the streamlet 4102 (576). The Q node 506 can send an acknowledgement message to the channel manager 214 indicating that the message has been stored successfully.
Here, a failed node can also be the node storing the head copy or the tail copy of the chain of copies. For example, if the Q node 208 fails, the channel manager 214 can instruct the MX node 204 first to send the message to the Q node 502 for storage in the copy #2 of the streamlet 4102. The message is then forwarded to the next adjacent copy in the chain for storage, until the message is stored in the tail copy.
If the Q node 506 fails, the channel manager 214 can repair the chain of copies of the streamlet 4102 such that the copy #3 on the Q node 504 becomes the tail copy of the chain. A message is first stored in the copy #1 on the Q node 208, then subsequently stored in the copy #2 on the Q node 502, and the copy #3 on the Q node 504. The Q node 504 then can send an acknowledgement message to the channel manager 214 indicating that the message has been stored successfully.
FIG. 5D is a data flow diagram of an example method 580 for repairing a chain of copies of a streamlet in the message system 100. In FIG. 5D, the channel manager 214 replaces the failed Q node 504 by allocating a new Q node 508 to store a copy #5 of the chain of copies of the streamlet 4102. In this example, the channel manager 214 instructs the MX node 204 to send a message from the publishers 406 to the Q node 208 for storage in the copy #1 of the streamlet 4102 (582). The message is then forwarded to the Q node 502 for storage in the copy #2 of the streamlet 4102 (584). The message is then forwarded to the Q node 508 for storage in the copy #5 of the streamlet 4012 (586). The message is then forwarded to the Q node 506 for storage in the copy #4 of the streamlet 4102 (588). The Q node 506 can send an acknowledgement message to the channel manager 214 indicating that the message has been stored successfully.
FIG. 6 is a flowchart of an example method 600 for publishing and subscribing messages for publishers and subscribers, respectively, of the system 100. The method can be implemented using, for example, a channel manager (e.g., channel manager 214) of the messaging system 100. The method begins by maintaining a plurality of channels, each channel including ordered messages stored in one or more buffers, each buffer having a respective predetermined time-to-live and copies residing on respective nodes (block 602). The method detects a failure of a first node of the plurality of nodes (block 604), wherein the first node stores respective first copies of first buffers for one or more first channels, and wherein at least one subscriber is subscribed to messages of the one or more first channels. For a first buffer with a first copy residing on the first node, the method identifies any other nodes as a second node storing a second copy of the first buffer (block 606). The method redirects the at least one subscriber and/or at least one publisher from the first node to the second node (block 608). The method retrieves messages stored in the second copy on the second node for the at least one subscriber (block 610).
Embodiments of the subject matter and the operations described in this specification can be implemented in digital electronic circuitry, or in computer software, firmware, or hardware, including the structures disclosed in this specification and their structural equivalents, or in combinations of one or more of them. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions, encoded on computer storage medium for execution by, or to control the operation of, data processing apparatus. Alternatively or in addition, the program instructions can be encoded on an artificially-generated propagated signal, e.g., a machine-generated electrical, optical, or electromagnetic signal that is generated to encode information for transmission to suitable receiver apparatus for execution by a data processing apparatus. A computer storage medium can be, or be included in, a computer-readable storage device, a computer-readable storage substrate, a random or serial access memory array or device, or a combination of one or more of them. Moreover, while a computer storage medium is not a propagated signal, a computer storage medium can be a source or destination of computer program instructions encoded in an artificially-generated propagated signal. The computer storage medium can also be, or be included in, one or more separate physical components or media (e.g., multiple CDs, disks, or other storage devices).
The operations described in this specification can be implemented as operations performed by a data processing apparatus on data stored on one or more computer-readable storage devices or received from other sources.
The term “data processing apparatus” encompasses all kinds of apparatus, devices, and machines for processing data, including by way of example a programmable processor, a computer, a system on a chip, or multiple ones, or combinations, of the foregoing. The apparatus can include special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit). The apparatus can also include, in addition to hardware, code that creates an execution environment for the computer program in question, e.g., code that constitutes processor firmware, a protocol stack, a database management system, an operating system, a cross-platform runtime environment, a virtual machine, or a combination of one or more of them. The apparatus and execution environment can realize various different computing model infrastructures, such as web services, distributed computing and grid computing infrastructures.
A computer program (also known as a program, software, software application, script, or code) can be written in any form of programming language, including compiled or interpreted languages, declarative, procedural, or functional languages, and it can be deployed in any form, including as a stand-alone program or as a module, component, subroutine, object, or other unit suitable for use in a computing environment. A computer program may, but need not, correspond to a file in a file system. A program can be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language resource), in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code). A computer program can be deployed to be executed on one computer or on multiple computers that are located at one site or distributed across multiple sites and interconnected by a communication network.
The processes and logic flows described in this specification can be performed by one or more programmable processors executing one or more computer programs to perform actions by operating on input data and generating output. The processes and logic flows can also be performed by, and apparatus can also be implemented as, special purpose logic circuitry, e.g., an FPGA (field programmable gate array) or an ASIC (application-specific integrated circuit).
Processors suitable for the execution of a computer program include, by way of example, both general and special purpose microprocessors, and any one or more processors of any kind of digital computer. Generally, a processor will receive instructions and data from a read-only memory or a random access memory or both. The essential elements of a computer are a processor for performing actions in accordance with instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include, or be operatively coupled to receive data from or transfer data to, or both, one or more mass storage devices for storing data, e.g., magnetic disks, magneto-optical disks, optical disks, or solid state drives. However, a computer need not have such devices. Moreover, a computer can be embedded in another device, e.g., a smart phone, a mobile audio or video player, a game console, a Global Positioning System (GPS) receiver, or a portable storage device (e.g., a universal serial bus (USB) flash drive), to name just a few. Devices suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including, by way of example, semiconductor memory devices, e.g., EPROM, EEPROM, and flash memory devices; magnetic disks, e.g., internal hard disks or removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and the memory can be supplemented by, or incorporated in, special purpose logic circuitry.
To provide for interaction with a user, embodiments of the subject matter described in this specification can be implemented on a computer having a display device, e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user and a keyboard and a pointing device, e.g., a mouse, a trackball, a touchpad, or a stylus, by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user can be received in any form, including acoustic, speech, or tactile input. In addition, a computer can interact with a user by sending resources to and receiving resources from a device that is used by the user; for example, by sending web pages to a web browser on a user's client device in response to requests received from the web browser.
Embodiments of the subject matter described in this specification can be implemented in a computing system that includes a back-end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front-end component, e.g., a client computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the subject matter described in this specification, or any combination of one or more such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include a local area network (“LAN”) and a wide area network (“WAN”), an inter-network (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks).
The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. In some embodiments, a server transmits data (e.g., an HTML page) to a client device (e.g., for purposes of displaying data to and receiving user input from a user interacting with the client device). Data generated at the client device (e.g., a result of the user interaction) can be received from the client device at the server.
A system of one or more computers can be configured to perform particular operations or actions by virtue of having software, firmware, hardware, or a combination of them installed on the system that in operation causes or cause the system to perform the actions. One or more computer programs can be configured to perform particular operations or actions by virtue of including instructions that, when executed by data processing apparatus, cause the apparatus to perform the actions.
While this specification contains many specific implementation details, these should not be construed as limitations on the scope of any inventions or of what may be claimed, but rather as descriptions of features specific to particular embodiments of particular inventions. Certain features that are described in this specification in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, various features that are described in the context of a single embodiment can also be implemented in multiple embodiments separately or in any suitable subcombination. Moreover, although features may be described above as acting in certain combinations and even initially claimed as such, one or more features from a claimed combination can in some cases be excised from the combination, and the claimed combination may be directed to a subcombination or variation of a subcombination.
Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous. Moreover, the separation of various system components in the embodiments described above should not be understood as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.
Thus, particular embodiments of the subject matter have been described. Other embodiments are within the scope of the following claims. In some cases, the actions recited in the claims can be performed in a different order and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In certain implementations, multitasking and parallel processing may be advantageous.

Claims

What is claimed is:

1. A method, comprising:

maintaining a plurality of channels,

wherein each channel comprises a plurality of messages stored in one or more buffers,

wherein a first node stores a first copy of a first buffer for a first channel of the plurality of channels,

wherein each of a plurality of secondary nodes stores a respective second copy of the first buffer, and

wherein at least one subscriber is subscribed to messages of the first channel;

detecting, by one or more computer processors, a failure of the first node;

identifying, by the one or more computer processors, one of the plurality of secondary nodes as a second node to access a second copy of the first buffer;

redirecting, by the one or more computer processors, the at least one subscriber from the first node to the second node; and

retrieving messages stored in the second copy on the second node for the at least one subscriber.

2. The method of claim 1, wherein the messages for each channel are stored in the one or more buffers according to an order.

3. The method of claim 1, wherein each buffer comprises a respective time-to-live.

4. The method of claim 1, wherein identifying one of the plurality of secondary nodes as the second node to access the second copy of the first buffer comprises:

selecting the second node based on a respective workload for each node of the plurality of secondary nodes.

5. The method of claim 1, comprising:

redirecting at least one publisher from the first node to the second node; and

sending messages from the at least one publisher to the second node for storage in the second copy.

6. The method of claim 1, wherein the respective second copies of the first buffer comprise a chain of copies including at least a head copy and a tail copy.

7. The method of claim 6, comprising:

forwarding messages stored in a particular copy of the chain of copies to a next adjacent copy of the chain of copies for storage, until messages in the head copy of the chain of copies are stored in the tail copy of the chain of copies.

8. The method of claim 1, wherein detecting the failure of the first node comprises:

detecting failure of the at least one subscriber obtaining messages from the first copy.

9. The method of claim 1, wherein at least one publisher sends messages of the first buffer to the first node for storage in the first copy before being redirected to the second node.

10. The method of claim 1, wherein detecting the failure of the first node comprises:

detecting failure of at least one publisher sending messages to the first node for storage in the first copy.

11. A system, comprising:

one or more computer processors programmed to perform operations to:

maintain a plurality of channels,

wherein at least one subscriber is subscribed to messages of the first channel;

detect a failure of the first node;

identify one of the plurality of secondary nodes as a second node to access a second copy of the first buffer;

redirect the at least one subscriber from the first node to the second node; and

retrieve messages stored in the second copy on the second node for the at least one subscriber.

12. The system of claim 11, wherein the messages for each channel are stored in the one or more buffers according to an order.

13. The system of claim 11, wherein to identify one of the plurality of secondary nodes as the second node to access the second copy of the first buffer the one or more computer processors are further to:

select the second node based on a respective workload for each node of the plurality of secondary nodes.

14. The system of claim 11, wherein the operations are further to:

redirect at least one publisher from the first node to the second node; and

send messages from the at least one publisher to the second node for storage in the second copy.

15. The system of claim 11, wherein the respective second copies of the first buffer comprise a chain of copies including at least a head copy and a tail copy.

16. The system of claim 15, wherein the operations are further to:

forward messages stored in a particular copy of the chain of copies to a next adjacent copy of the chain of copies for storage, until messages in the head copy of the chain of copies are stored in the tail copy of the chain of copies.

17. The system of claim 11, wherein to detect the failure of the first node the one or more computer processors are further to:

detect failure of the at least one subscriber obtaining messages from the first copy.

18. The system of claim 11, wherein at least one publisher sends messages of the first buffer to the first node for storage in the first copy before being redirected to the second node.

19. The system of claim 11, wherein to detect the failure of the first node the one or more computer processors are further to:

detect failure of at least one publisher sending messages to the first node for storage in the first copy.

20. A non-transitory computer-readable medium having instructions stored thereon that, when executed by one or more computer processors, cause the one or more computer processors to:

maintain a plurality of channels,

wherein at least one subscriber is subscribed to messages of the first channel;

detect a failure of the first node;