This document collects and organizes thoughts about the design of the Gerrit multi-site plugin, supporting the definition of the implementation roadmap.
It first presents background for the problems the plugin will address and the tools currently available in the Gerrit ecosystem that support the solution. It then lays out an overall roadmap for implementing support for Gerrit multi-site, and a snapshot of the current status of the design including associated limitations and constraints.
Companies that adopt Gerrit as the center of their development and review pipeline often have the requirement to be available on a 24/7 basis. This requirement may extend across large and geographically distributed teams in different continents.
Because of constraints defined by the CAP theorem designing a performant and scalable solution is a real challenge.
Vertical scaling is one of the options to support a high load and a large number of users. A powerful server with multiple cores and sufficient RAM to potentially fit the most frequently used repositories simplifies the design and implementation of the system. The relatively reasonable cost of hardware and availability of multi-core CPUs make this solution highly attractive to some large Gerrit setups. Further, the central Gerrit server can be duplicated with an active/passive or active/active high availability configuration with the storage of the Git repositories shared across nodes through dedicated fibre-channel lines or SANs.
This approach can be suitable for mid to large-sized Gerrit installations where teams are co-located or connected via high-speed dedicated networks. However, when teams are located on opposite sides of the planet, even at the speed of light the highest theoretical fire-channel direct connection can be limiting. For example, from San Francisco to Bangalore the theoretical absolute minimum latency is 50 msec. In practice, however, it is often around 150/200 msec in the best case scenarios.
In the alternate option, horizontal scaling, the workload is spread across several nodes, which are distributed to different locations. For our teams in San Francisco and Bangalore, each accesses a set of Gerrit masters located closer to their geographical location, with higher bandwidth and lower latency. (To control operational cost from the proliferation of servers, the number of Gerrit masters can be scaled up and down on demand.)
This solution offers a higher level of scalability and lower latency across locations, but it requires a more complex design.
The vertical and horizontal approaches can be combined to achieve both high performance on the same location and low latency across geographically distributed sites.
Geographical locations with larger teams and projects can have a bigger Gerrit server in a high availability configuration, while locations with less critical service levels can use a lower-spec setup.
The multi-site plugin enables the OpenSource version of Gerrit Code Review to support horizontal scalability across sites.
Gerrit has already been deployed in a multi-site configuration at Google and in a multi-master fashion at Qualcomm. Both implementations include fixes and extensions that are tailored to the specific infrastructure requirements of each company‘s global networks. Those solutions may or may not be shared with the rest of the OpenSource Community. Specifically, Google’s deployment is proprietary and not suitable for any environment outside Google‘s data-centers. Further, in Qualcomm’s case, their version of Gerrit is a fork of v2.7.
In contrast, the multi-site plugin is based on standard OpenSource components and is deployed on a standard cloud environment. It is currently used in a multi- master and multi-site deployment on GerritHub.io, serving two continents (Europe and North America) in a high availability setup on each site.
The development of the multi-site support for Gerrit is complex and thus has been deliberately broken down into incremental steps. The starting point is a single Gerrit master deployment, and the end goal is a fully distributed set of cooperating Gerrit masters across the globe.
The transition between steps requires not only an evolution of the Gerrit setup and the set of plugins but also the implementation of more mature methods to provision, maintain and version servers across the network. Qualcomm has pointed out that the evolution of the company culture and the ability to consistently version and provision the different server environments are winning features of their multi-master setup.
Google is currently running at Stage #10. Qualcomm is at Stage #4 with the difference that both masters are serving RW traffic, which is possible because the specifics of their underlying storage, NFS and JGit implementation allows concurrent locking at the filesystem level. GerritHub is running at Stage #9, with 3 locations.
Having all the repositories replicated to all sites could be, in some cases, not a great idea. The rationale can be explained with a simple example.
Company FooCompany is developing a new huge and secret project code-named Tango with a software engineering team all geo-located in India. The Git repository is huge and contains millions of refs and packfiles for tens of GBytes. Project Tango requires also to have some medium-sizes binaries in the Git repository. FooCompany has a multi-site deployment across the globe, covering Europe, USA, Australia and China, other than India, where the new project is developed.
The teams in Europe and USA are involved in the project, from a code-review perspective. Their engineers are typically using the Gerrit UI for reviews and fetch individual patch-sets for local verification.
All projects are replicated everywhere, including the Tango project. The replication creates a huge network overload across the globe.
When an engineer is pushing a packfile in India, it gets replicated to all sites, causing congestion on the replication channel. When a software engineer in Europe reviews the changes of the Tango project, it creates modifications to the NoteDb meta ref that would be then replicated back to India with a non-neglibigle latency, due to the size of the repository and the huge refs advertisement phase implied in the replication.
Software engineers around the globe do not need to see the Tango project, with the exception of the reviewers in Europe and USA. However, everyone is impacted and the servers and replication channels are overloaded.
The multi-site setup is using a sharding logic, projects are replicated or not depending on how they are classified:
The Tango project is a local project because it is mainly developed in one site: India.
When an engineer is pushing a packfile in India, it does not get replicated to all sites, saving bandwidth for the global projects replication. When a software engineer in Europe opens a change associated with the Tango project, he gets silently redirected to the site in India where the project is located.
Reviewers commenting on changes of the Tango project, create modifications to the NoteDb in India, which are immediately visible to the local software engineers, without a long replication lag.
Software engineers around the globe do not need to see the Tango project, with the exception of the reviewers in Europe and USA. The Tango project is not visible and not replicated to the other sites and, the people not involved in the project, are not impacted at all.
Consider also pull replication for cases like 5, 6, 7... which could be done also synchronously to the incoming write operation. In case a write operation fails to be replicated by the master node(s), it could be automatically rolled back and reported to the client for retry. This would provide 100% loss-less disaster recovery support.
When running pull replication asynchronously, similarly to the replication plugin, an unrecoverable crash of the replication source would result in unnoticed data loss. The only way to recover the data would be telling the users who pushed the commits to push them again. However, someone needs to manually detect the issue in the replication log and get in touch with the user.
The pull-replication plugin supports synchronous replication and has the structure to perform also the asynchronous variant in the future.
This plugin expands upon the excellent work on the high-availability plugin, introduced by Ericsson for implementing mutli-master at Stage #4. The git log history of this projects still shows the ‘branching point’ where it started. The v2.16.x (with NoteDb) of the multi-site plugin was at Stage #7.
The current version of the multi-site plugin is at Stage #9, it is now possible for Gerrit data to be available in two or more separate geo-locations (e.g. San Francisco, Frankfurt and Bangalore), each serving local traffic through the local instances with minimum latency.
You may be questioning the reasoning behind creating yet another plugin for multi-master, instead of maintaining a single code-base with the high-availability plugin. The choice stems from the differing design considerations to address scalabiilty, as discussed above for the vertical (single site) and horizonal (multi-site) approaches.
In theory, one could keep a single code-base to manage both approaches, however the result would be very complicated and difficult to configure and install. Having two more focussed plugins, one for high availability and another for multi-site, allows us to have a simpler, more usable experience, both for developers of the plugin and for the Gerrit administrators using it.
The high-availability and multi-site plugins are solutions to different problems. Two or more nodes on the same site are typically deployed to increase the reliability and scalability of a Gerrit setup, however, doesn't provide any benefit in terms of data access across locations. Replicating the repositories to remote locations does not help the scalability of a Gerrit setup but is more focused on reducing the data transfer time between the client and the server, thanks to the higher bandwidth available in the local regions.
There are some advantages in implementing multi-site at Stage #9:
Optimal latency of the Git read/write operations on all sites, and signficant improvement of the Gerrit UI responsiveness, thanks fo the reduction of the network latency.
High SLA (99.99% or higher, source: GerritHub.io) can be achieved by implementing network distribution across sites.
Access transparency through a single Gerrit URL, thanks to a geo-location DNS routing.
Automatic failover, disaster recovery, and failover to remote sites.
All sites have local consistency, with the assurance of global eventual consistency.
The current limitations of Stage #9 are:
Limited supports for many sites: One could, potentially, support a very high number of sites, but the pull-replication logic to all sites could have a serious consequence in the overall perceived latency. Having to deal with a very high number of site requires the implementation of a quorum on all the nodes available for replication.
Requires Gerrit v3.0 or later: Data conisistency requires a server completely based on NoteDb. If you are not familiar with NoteDb, please read the relevant section in the Gerrit documentation.
Let's suppose you have two sites, in San Francisco and Bangalore. The modifications of data will flow from San Francisco to Bangalore and the other way round.
Depending on the network infrastructure between the two sites latency can range between seconds and minutes. The available bandwith is low, so the Gerrit admin decides to use a traditional push replication (asynchronous) between the two sites.
When a developer located in Bangalore accesses a repository for which most pushes originate from San Francisco, he may see a “snapshot in the past” of the data, both from the Gerrit UI and on the Git repository served locally. In contrast, a developer located in San Francisco will always see on his repository the “latest and greatest” of everything. Things are exactly in the other way around for a repository that is mainly receiving pushes from developers in Bangalore.
Should the central site in San Francisco become unavailable for a significant period of time, the Bangalore site will still be able to serve all Gerrit repositories, including those where most pushes come from San Francisco. People in San Francisco can't access their local site anymore, because it is unavailable. All the Git and Gerrit UI requests will be served remotely by the Bangalore server while the local system is down. When the San Francisco site comes up again, and passes the “necessary checks”, it will become the main site again for the users in the same geo location..
This section goes into the high-level design of the current solution, lists the components involved, and describes how the components interact with each other.
There are several distinct classes of information that have to be kept consistent across different sites in order to guarantee seamless operation of the distributed system.
Git repositories: They are stored on disk and are the most important information to maintain. The repositories store the following data:
Git BLOBs, objects, refs and trees.
NoteDb, including Groups, Accounts and review data
Project configurations and ACLs
Project submit rules
Indexes: A series of secondary indexes to allow search and quick access to the Git repository data. Indexes are persistent across restarts.
Caches: A set of in-memory and persisted data designed to reduce CPU and disk utilization and to improve performance.
Web Sessions: Define an active user session with Gerrit, used to reduce load to the underlying authentication system. Sessions are stored by default on the local filesystem in an H2 table but can be externalized via plugins, like the WebSession Flatfile.
To achieve a Stage #9 multi-site configuration, all the above information must be replicated transparently across sites.
The multi-site solution described here depends upon the combined use of different components:
multi-site libModule: exports interfaces as DynamicItems to plug in specific implementation of
Global Ref-DB plugins.
broker plugin: an implementation of the broker interface, which enables the replication of Gerrit indexes, caches, and stream events across sites. When no specific implementation is provided, then the Broker Noop implementation then libModule interfaces are mapped to internal no-ops implementations.
Global Ref-DB plugin: an implementation of the Global Ref-DB interface, which enables the detection of out-of-sync refs across gerrit sites. When no specific implementation is provided, then the Global Ref-DB Noop implementation then libModule interfaces are mapped to internal no-ops implementations.
replication plugin: enables asynchronous push replication of the Git repositories across sites.
pull replication plugin: enables the synchronous replication of the Git repositories across sites.
web-session broker plugin: supports the storage of active sessions to a message broker topic, which is then broadcasted across sites.
health check plugin: supports the automatic election of the RW site based on a number of underlying conditions of the data and the systems.
HA Proxy: provides the single entry-point to all Gerrit functionality across sites.
The interactions between these components are illustrated in the following diagram:
As mentioned earlier there are different components behind the overarching architecture of this solution of a distributed multi-site gerrit installation, each one fulfilling a specific goal. However, whilst the goal of each component is well-defined, the mechanics on how each single component achieves that goal is not: the choice of which specific message broker or which Ref-DB to use can depend on different factors, such as scalability, maintainability, business standards and costs, to name a few.
For this reason the multi-site component is designed to be explicitly agnostic to specific choices of brokers and Global Ref-DB implementations, and it does not care how they, specifically, fulfill their task.
Instead, this component takes on only two responsibilities:
Wrapping the GitRepositoryManager so that every interaction with git can be verified by the Global Ref-DB plugin.
Exposing DynamicItem bindings onto which concrete Broker and a Global Ref-DB plugins can register their specific implementations. When no such plugins are installed, then the initial binding points to no-ops.
Detect out-of-sync refs across multiple gerrit sites: Each change attempting to mutate a ref will be checked against the Ref-DB to guarantee that each node has an up-to-date view of the repository state.
Each gerrit node in the cluster needs to be informed and inform all other nodes about fundamental events, such as indexing of new changes, cache evictions and stream events. This component will provide a specific pub/sub broker implementation that is able to do so.
When provided, the message broker plugin will override the dynamicItem binding exposed by the multi-site module with a specific implementation, such as Kafka, RabbitMQ, NATS, etc.
Noop implementation provided by the
Multi-site libModule does nothing upon publishing and producing events. This is useful for setting up a test environment and allows multi-site library to be installed independently from any additional plugins or the existence of a specific broker installation. The Noop implementation can also be useful when there is no need for coordination with remote nodes, since it avoids maintaining an external broker altogether: for example, using the multi-site plugin purely for the purpose of replicating the Git repository to a disaster-recovery site and nothing else.
Whilst the replication plugin allows the propagation of the Git repositories across sites and the broker plugin provides a mechanism to propagate events, the Global Ref-DB ensures correct alignment of refs of the multi-site nodes.
It is the responsibility of this plugin to store atomically key/pairs of refs in order to allow the libModule to detect out-of-sync refs across multi sites. (aka split brain). This is achieved by storing the most recent
sha for each specific mutable
refs, by the usage of some sort of atomic Compare and Set operation.
We mentioned earlier the CAP theorem, which in a nutshell states that a distributed system can only provide two of these three properties: Consistency, Availability and Partition tolerance: the Global Ref-DB helps achieving Consistency and Partition tolerance (thus sacrificing Availability).
See Prevent split brain thanks to Global Ref-DB For a thorough example on this.
When provided, the Global Ref-DB plugin will override the dynamicItem binding exposed by the multi-site module with a specific implementation, such as Zoekeeper, etcd, MySQL, Mongo, etc.
Noop implementation provided by the
Multi-site libModule accepts any refs without checking for consistency. This is useful for setting up a test environment and allows multi-site library to be installed independently from any additional plugins or the existence of a specific Ref-DB installation.
The replication of the Git repositories, indexes, cache and stream events happen on different channels and at different speeds. Git data is typically larger than meta-data and has higher latency than reindexing, cache evictions or stream events. This means that when someone pushes a new change to Gerrit on one site, the Git data (commits, BLOBs, trees, and refs) may arrive later than the associated reindexing or cache eviction events.
It is, therefore, necessary to handle the lack of synchronization of those channels in the multi-site plugin and reconcile the events at the destinations.
The solution adopted by the multi-site plugin supports eventual consistency at rest at the data level, thanks to the following two components which:
Identify not-yet-processable events: A mechanism to recognize not-yet-processable events related to data not yet available (based on the timestamp information available on both the metadata update and the data event)
Queue not-yet-processable events: A queue of not-yet-processable events and an asynchronous processor to check if they became processable. The system also is configured to discard events that have been in the queue for too long.
Stream events are wrapped into an event header containing a source identifier. Events originated by the same node in the broker-based channel are silently dropped so that they do not replicate multiple times.
Gerrit has the concept of server-id which, unfortunately, does not help solve this problem because all the nodes in a Gerrit cluster must have the same server-id to allow interoperability of the data stored in NoteDb.
The multi-site plugin introduces a new concept of instance-id, which is a UUID generated during startup and saved into the data folder of the Gerrit site. If the Gerrit site is cleared or removed, a new id is generated and the multi-site plugin will start consuming all events that have been previously produced.
The concept of the instance-id is very useful. Since other plugins could benefit from it, it will be the first candidate to move into the Gerrit core, generated and maintained with the rest of the configuration. Then it can be included in all stream events, at which time the multi-site plugin's “enveloping of events” will become redundant.
The broker based solutions improve the resilience and scalability of the system. But there is still a point of failure: the availability of the broker itself. However, using the broker does allow having a high-level of redundancy and a multi-master / multi-site configuration at the transport and storage level.
At the moment, the acknowledge level for publication can be controlled via configuration and allows to tune the QoS of the publication process. Failures are explicitly not handled at the moment; they are just logged as errors. There is no retry mechanism to handle temporary failures.
The current solution of multi-site at Stage #7 with asynchronous replication risks that the system will reach a Split Brain situation (see issue #10554).
In this case we are considering two different clients each doing a
push on top of the same reference. This could be a new commit in a branch or the change of an existing commit.
t0: both clients see the status of
Instance1 is the RW node and will receive any
Instance2 are in sync at
W1. The request is served by
Instance1 which acknowledges it and starts the replication process (with some delay).
t2: The replication operation is completed. Both instances are in a consistent state
W0 -> W1.
Client1 shares that state but
Client2 is still behind.
W2 which is still based on
W0 -> W2). The request is served by
Instance2 which detects that the client push operation was based on an out-of-date starting state for the ref. The operation is refused.
Client2 synchronises its local state (e.g. rebases its commit) and pushes
W0 -> W1 -> W2. That operation is now considered valid, acknowledged and put in the replication queue until
Instance1 becomes available.
Instance1 restarts and is replicated at
W0 -> W1 -> W2
In this case the steps are very similar except that
Instance1 fails after acknowledging the push of
W0 -> W1 but before having replicated the status to
W0 -> W2 to
Instance2, this is considered a valid operation. It gets acknowledged and inserted in the replication queue.
Instance1 restarts. At this point both instances have pending replication operations. They are executed in parallel and they bring the system to divergence.
Root causes of the Split Brain problem:
pushoperation before all replicas are fully in sync.
Two possible approaches to solve the Split Brain problem:
Synchronous replication: In this case the system would behave essentially as the happy path diagram shown above and would solve the problem by operating on the first of the causes, at the expense of performance, availability and scalability. It is a viable and simple solution for two nodes set up with an infrastructure allowing fast replication.
Centralise the information about the latest status of mutable refs: This would operate on the second cause. That is, it would allow instances to realise that they are not in sync on a particular ref and refuse any write operation on that ref. The system could operate normally on any other ref and also would impose no limitation in other functions such as, Serving the GUI, supporting reads, accepting new changes or patch-sets on existing changes. This option is discussed in further detail below.
NOTE: The two options are not exclusive.
The above scenario can be prevented by using an implementation of the Global Ref-DB interface, which will operate as follows:
The difference, in respect to the split brain use case, is that now, whenever a change of a mutable ref is requested, the Gerrit server verifies with the central RefDB that its status for this ref is consistent with the latest cluster status. If that is true the operation succeeds. The ref status is atomically compared and set to the new status to prevent race conditions.
In this case
Instance2 enters a Read Only mode for the specific branch until the replication from
Instance1 is completed successfully. At this point write operations on the reference can be recovered. If
Client2 can perform the
push again vs
Instance2, the server recognises that the client status needs an update, the client will
push the correct status.
NOTE: This implementation will prevent the cluster to enter split brain but might result in a set of refs in Read Only state across all the cluster if the RW node is failing after having sent the request to the Ref-DB but before persisting this request into its
Once you go multi-site multi-master you can improve the latency of your calls by serving traffic from the closest server to your user.
Whether you are running your infrastructure in the cloud or on premise you have different solutions you can look at.
Route53 AWS DNS service offers the opportunity of doing Geo Proximity routing using Traffic Flow.
Traffic flow is a tool which allows the definition of traffic policies and rules via a UI. Traffic rules are of different types, among which Geoproximity rules.
When creating geoproximity rules for your resources you can specify one of the following values for each rule:
This allows you to have an hybrid cloud-on premise infrastructure.
You can define quite complex failover rules to ensure high availability of your system (here an example).
Overall the service provided is pretty much a smart reverse-proxy, if you want more complex routing strategies you will still need a proper Load Balancer.
GCE doesn't offer a Geographical based routing, but it implicitly has geo-located DNS entries when distributing your application among different zones.
The Load Balancer will balance the traffic to the nearest available instance , but this is not configurable and the app server has to be in GC.
Hybrid architectures are supported but would make things more complicated, hence this solution is probably worthy only when the Gerrit instances are running in GC.
If you are going for an on premise solution and using HAProxy as Load Balancer, it is easy to define static ACL based on IP ranges and use them to route your traffic.
This blogpost explains how to achieve it.
On top of that, you want to define a DNS entry per zone and use the ACLs you just defined to issue redirection of the calls to most appropiate zone.
You will have to add to your frontend definition your redirection strategy, i.e.:
http-request redirect code 307 prefix https://review-eu.gerrithub.io if acl_EU http-request redirect code 307 prefix https://review-am.gerrithub.io if acl_NA
Auto-reconfigure HAProxy rules based on the projects sharding policy
Implement more global-refdb storage layers (e.g. TiKV) and more cloud-native message brokers (e.g. NATS)
Implement a quorum-based policy for accepting or rejecting changes in the pull-replication plugin
Allow asynchronous pull-replication across sites, based on asynchronous events through the message broker
Implement a “fast replication path” for NoteDb-only changes, instead of relying on the Git protocol