Full source documentation can be found at https://derecho-project.github.io/.
This is the main repository for the Derecho project. It contains all of the Derecho library code, as well as several sample applications and test programs.
- Intended use cases and assumptions
- Organization
- Installation
- Using Derecho
- Notes on Very Large Deployments
Derecho is aimed at supporting what are called "cloud micro-services", meaning pools of servers that would reside in a cluster or on a cloud data-center, probably to support the "first tier" where requests arrive from external clients, and that perform some kind of well-defined subtask like data cleaning, image classification, compression and deduplication, etc. Although the system could definitely run in other settings, you may need help if you try to port the system to a setting we have never tested it on.
The functionality of Derecho centers on:
- Forming a "group" of servers ("processes"). Membership is automatically tracked and changes are reported via upcalls to your code.
- Structuring your server group into subgroups. You would typically have one subgroup for each distinct functionality used. For example, suppose you were implementing a storage service similar to Apache's Hadoop file system (HDFS). HDFS has a name-node service, a meta-data service and a storage service. In Derecho, each would be implemented by a subgroup. Subgroups can have distinct or overlapping membership, under your control. For scalability, a subgroup would typically be sharded into even smaller subgroups. For example, a storage service might have an "array" of mini-storage servers, each handling a subset of the files. There would be one shard per subset, and the members of that shard would replicate the identical contents. Unlike in Paxos-based systems you may have read about elsewhere, Derecho supports a full replication model: every active (non-failed) replica has identical, complete contents.
- Automated repair after crashes, recoveries, and automated initialization (from a checkpoint) when a new member joins.
- Automated persisted storage (append-only logging) if desired.
Derecho is intended to achieve the fastest performance possible on a given platform. An aspect of this is to leverage modern RDMA hardware, which is becoming quite common. However, because not every system has RDMA at this time, Derecho can also be configured to run on high-speed, low-latency TCP links (a few other options exist as well, such as Intel OMNI Path). The system design does expect that server nodes have rather uniform properties, and that this also be true for the links between them. Derecho also assumes a relatively low rate of membership changes. Processes can join or leave, and we even batch joins and leaves if a burst of them occurs, but the design of the system really assumes long periods (think "minutes") of stability, and we basically lock everything down during the short periods (think "hundreds of milliseconds") needed for reconfiguration.
In Figure I of our paper (http://www.cs.cornell.edu/ken/derecho-tocs.pdf) you can see an example setup. Notice the external clients. You can have as many of those as you like, and they can connect in all sorts of ways: TCP, RESTful RPC, WCF -- whatever you feel comfortable with (you have to do that part yourself, so use tools familiar to you). But within the microservice itself, you want the members to be in a single cluster or datacenter, with high-speed low-latency links from node to node, no firewalls that would block our connections, etc.
Configured to run on TCP, Derecho should work when virtualized (true VMs or containers). With RDMA, you probably need bare-metal or some sort of container running on bare-metal. RDMA hardware doesn't virtualize well and we aren't interested in trying to get that to work. In the TCP case, you can run a few instances of Derecho on one machine, but don't push this to the point where they would be suffering big scheduling or paging delays: that isn't matched to our intended use case, and we won't be able to help if you try that and it doesn't work well.
To obtain such high speeds, Derecho sacrifices some of the interoperability seen with other platforms, such as the CORBA or JINI communication layers. Unlike those technologies, Derecho does not convert data into a universally-compatible wire format. Instead, it requires that the sender and receiver have identical endian formats, and identical data structure layouts and padding. The best way to ensure that this is so is to only build a Derecho group using members that will run on the same computer architecture (for example, i86 or ARM), and only build the code with the same version of C++, configured with the same language level and optimization level. We are implementing a self-test in Derecho to detect that an incompatible system is attempting to join a Derecho group, and reject the request. However, this is not yet part of the system.
Derecho does not have any specific O/S dependency. We've tested most extensively on the current release of Ubuntu, which is also the default Linux configuration used on Azure IoT Edge and Azure IoT, but there is no reason the system couldn't be used on other platforms.
This project is organized as a standard CMake-built C++ library: all headers are within the include/derecho/ directory, all CPP files are within the src/ directory, and each subdirectory within src/ contains its own CMakeLists.txt that is included by the root directory's CMakeLists.txt. Within the src/ and include/derecho/ directories, there is a subdirectory for each separate module of Derecho, such as RDMC, SST, and Persistence. Some sample applications and test cases that are built on the Derecho library are included in the src/applications/ directory, which is not included when building the Derecho library itself.
Derecho is a library that helps you build replicated, fault-tolerant services in a datacenter with RDMA networking. Here's how to start using it in your projects.
- You will start by verifying that you have a compatible operating system (we do our development on Ubuntu but CentOS should be fine) and network (we recommend RDMA or TCP).
- Next, decide if you prefer to use a pre-compiled container, which is easier, or would like to build from source.
- If building from source, you will clone the code base in the place you plan to create the release binaries, then follow the instructions for creating a folder in which the binaries will reside. Then from "prerequisites" run the .sh shell scripts one by one, for example "sudo ./install-foo.sh". These should complete without error messages -- if you get warnings or errors, stop and post a question about it on the discussions page.
- Next, if building from source, follow the remainder of the "installing Derecho" instructions.
- Last, clone and build the Cascade code base.
- You should now be able to set up a configuration file and run our demos.
Derecho was designed to run correctly on:
- RDMA, which can be deployed on Infiniband or Ethernet (RoCE).
- TCP (but not UDP).
Under the surface, the specific requirements are these. First, Derecho needs the network to be reliable, ordered, to deliver data exactly once in the order sent, to have its own built-in flow control, and to automatically detect failures. TCP works this way, and is widely available. RDMA mimics TCP and adds some features of its own, and we leverage those when we have access to them. For example, RDMA allows us to directly write into the memory of a remote machine, with permissions. This is very useful and is a core technology that Derecho was designed around.
Derecho accesses the network via a configurable layer that currently offers:
- Direct use of RDMA verbs.
- Indirect networking, via a wrapper called "LibFabric" that offers efficient mappings to TCP or RDMA verbs.
- Eventually (Derecho V2.3 or beyond): the DataPlane Developers Kit, DPDK. Again, our DPDK support will focus on mappings to TCP or RDMA.
Your job when configuring Derecho is to build with one of these choices (by default we use LibFabric), then tell us in the Derecho config file which "sub-choice" to make, if any. So, for LibFabric, you must tell us "tcp" or "verbs". Do not use the LibFabrics "socket" provider -- this is deprecated and can cause Derecho to malfunction.
It is quite easy to misconfigure Derecho. You can simply tell us to map to a protocol like UDP, or some other option that lacks the semantics shared by TCP and RDMA. If you do that, our system will break -- it may start up, but then it will crash.
How would you decide whether the LibFabrics or DPDK mapping to some other technology is safe for Derecho? Your task centers on understanding what we require, then verifying that your favorite technology matches our needs. Specifically, Derecho requires:
- A way to send data as a stream of blocks (one-sided RDMA remote writes), each operation telling the transport system where to put some block of data on the remote node. This is how RDMA works "out of the box." When we run on TCP, we depend on LibFabric to implement this abstraction.
- With each block of bytes, we expect our writes to be applied in the order we sent them, and we require "cache line atomicity" from the memory subsystem on the target machine. This just means that if we write 1234 into X, and then write 4567 into X (X denotes a variable inside one 64-byte cache line), a reader might see 1234 for a while, but then eventually would see 4567. The update order doesn't change, and the bytes don't somehow become mashed together.
- Memory fencing on an operation-by-operation basis. This means that if we do two writes, block A and then block B, and a remote process can see the data from B, it will also see the data from A. Memory fencing implies cache-coherency: no remote process that sees data from B would run into stale cached data from before A or B occurred.
- Sequential consistency for non-fenced updates from a single source. Derecho uses memory fencing -- not locking. Memory fencing is weaker than "atomics", which is a property needed if a system has multiple writers contending to update the same location. We don't need atomics: in Derecho, any given memory location has just a single writer.
You may be surprised to learn that TCP has these guarantees. In fact, TCP "on its own" only has some of them. The fencing property is really a side-effect of using LibFabric, which emulates RDMA on TCP in its tcp transport protocol for one-sided writes. In RDMA, the properties we described are standard.
Equally, it may be surprising that LibFabric on other transports, such as UDP, lacks this guarantee. In fact this puzzles us too: LibFabrics was originally proposed as a way to emulate RDMA on TCP and other transports, and you would expect that its one-sided RDMA write implementation really should be faithful to RDMA semantics. However, when LibFabric was extended to support UDP, the solution was designed to emulate RDMA to the extent possible, but without ever retransmitting bytes. Thus with UDP, LibFabric will not be reliable, although it will be sequentially consistent and separate verbs will be memory-fenced. This is not strong enough for Derecho.
Above, we noted that we hope to support Derecho on DPDK. DPDK, like LibFabrics, is a very thin standard wrapper over various user-space networking and remote memory options (including RDMA). DPDK passes the properties of the underlying transport through. DPDK does support a third-party layer, urdma, that translates RDMA verbs into DPDK operations, and we believe that this mix should work over TCP (so: urdma on DPDK on TCP looks plausible to us). URDMA on DPDK on RDMA looks fine too. But urdma on DPDK on UDP would not work because UDP is unreliable.
- Linux (other operating systems don't currently support the RDMA features we use)
- A C++ compiler supporting C++17: GCC 7.3+ or Clang 7+
- CMake 2.8.1 or newer
- The OpenSSL SSL/TLS Library. On Ubuntu and other Debian-like systems, you can install package
libssl-dev. We tested with v1.1.1f. But it should work for any version >= 1.1.1. - The "rdmacm" and "ibverbs" system libraries for Linux, at version 17.1 or higher. On Ubuntu and other Debian-like systems, these are in the packages
librdmacm-devandlibibverbs-dev. spdlog, a logging library, v1.3.1 or newer. On Ubuntu 19.04 and later this can be installed with the packagelibspdlog-dev. The version of spdlog in Ubuntu 18.04's repositories is too old, but if you are running Ubuntu 18.04 you can download thelibspdlog-devpackage here and install it manually with no other dependencies needed.- The Open Fabric Interface (OFI) library:
libfabric. Since this library's interface changes significantly between versions, please installv1.12.1from source rather than any packaged version. (Installation script) - Lohmann's JSON for Modern C++ library, v3.9 or newer. This library is not packaged for Ubuntu, but can easily be installed with our installation script.
- @mpmilano's C++ utilities, which are all CMake libraries that can be installed with "make install":
To download the project, run
git clone https://github.com/Derecho-Project/derecho.git
Once cloning is complete, to build the code, cd into the derecho directory and run:
mkdir Releasecd Releasecmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=<path-to-install-dir> ..make -j $(nproc)
This will place the binaries and libraries in the sub-directories of Release.
The other build type is Debug. If you need to build the Debug version, replace Release by Debug in the above instructions. We explicitly disable in-source build, so running cmake . in derecho will not work.
Once the project is built, install it by running:
make install
By default, Derecho will be installed into /usr/local/. Please make sure you have sudo privileges to write to system directories.
Successful installation will set up the followings in $DESTDIR:
include/derecho- the header fileslib/libderecho.so- the main shared librarylib/libdpods.so- the derecho Old-Plain-Data storage librarylib/cmake/derechoandlib/cmake/dpods- cmake support forfind_package(derecho)/find_package(dpods)share/derecho- sample derecho configuration files.
To uninstall, run:
rm -rf `cat install_manifest.txt`
To build your own derecho executable, simple run:
g++ -std=c++1z -o myapp myapp.cpp -lderecho -pthread
To use Derecho in your code, you simply need to
- include the header
derecho/core/derecho.hppin your *.h *.hpp or *.cpp files, and - specify a configuration file, either by setting environment variable
DERECHO_CONF_FILEor by placing a file namedderecho.cfgin the working directory. A sample configuration file along with an explanation can be found in<installation-prefix>/share/derecho/derecho-sample.cfg.
The configuration file consists of three sections: DERECHO, RDMA, and PERS. The DERECHO section includes core configuration options for a Derecho instance, which every application will need to customize. The RDMA section includes options for RDMA hardware specifications. The PERS section allows you to customize the persistent layer's behavior.
Applications need to tell the Derecho library which node is the initial leader with the options leader_ip and leader_gms_port. Each node then specifies its own ID (local_id) and the IP address and ports it will use for Derecho component services (local_ip, gms_port, state_transfer_port, sst_port, and rdmc_port). Also, if using external clients, applications need to specify the ports serving external clients (external_port);
The other important parameters are the message sizes. Since Derecho pre-allocates buffers for RDMA communication, each application should decide on an optimal buffer size based on the amount of data it expects to send at once. If the buffer size is much larger than the messages an application actually sends, Derecho will pin a lot of memory and leave it underutilized. If the buffer size is smaller than the application's actual message size, it will have to split messages into segments before sending them, causing unnecessary overhead.
Three message-size options control the memory footprint and performance of Derecho. In all cases, larger values will increase the memory (DRAM) footprint of the application, and it is fairly easy to end up with a huge memory size if you just pick giant values. The defaults keep the memory size smaller, but can reduce performance if an application is sending high rates of larger messages.
The options are named max_payload_size, max_smc_payload_size, block_size, max_p2p_request_payload_size, and max_p2p_reply_payload_size.
No message bigger than max_payload_size will be sent by Derecho multicast(derecho::Replicated::send). No message bigger than max_p2p_request_payload_size will be sent by Derecho p2p send(derecho::Replicated::p2p_send or derecho::ExternalClientCaller::p2p_send). No reply bigger than max_p2p_reply_payload_size will be sent to carry the return values any multicast or p2p send.
To understand the other two options, it helps to remember that internally, Derecho makes use of two sub-protocols when it transmits your data. One sub-protocol is optimized for small messages, and is called SMC. Messages equal to or smaller than max_smc_payload_size will be sent using SMC. Normally max_smc_payload_size is set to a small value, like 1K, but we have tested with values up to 10K. This limit should not be made much larger: performance will suffer and memory would bloat.
Larger messages are sent via RDMC, our big object protocol. These will be automatically broken into chunks. Each chunk will be of size block_size. The block_size value we tend to favor in our tests is 1MB, but we have run experiments with values as large as 100MB. If you plan to send huge objects, like 100MB or even multi-gigabyte images, consider a larger block size: it pays off at that scale. If you expect that huge objects would be rare, use a value like 1MB.
More information about Derecho parameter setting can be found in the comments in the default configuration file. You may want to read about window_size, timeout_ms, and rdmc_send_algorithm.
The most important configuration entries in this section are provider and domain. The provider option specifies the type of RDMA device (i.e. a class of hardware) and the domain option specifies the device (i.e. a specific NIC or network interface). This Libfabric document explains the details of those concepts.
The tx_depth and rx_depth configure the maximum of pending requests that can be waiting for acknowledgement before communication blocks. Those numbers can be different from one device to another. We recommend setting them as large as possible.
Here are some sample configurations showing how Derecho might be configured for two common types of hardware.
Configuration 1: run Derecho over TCP/IP with Ethernet interface 'eth0':
...
[RDMA]
provider = sockets
domain = eth0
tx_depth = 256
rx_depth = 256
...
Configuration 2: run Derecho over verbs RDMA with RDMA device 'mlx5_0':
...
[RDMA]
provider = verbs
domain = mlx5_0
tx_depth = 4096
rx_depth = 4096
...
The application can specify the location for persistent state in the file system with file_path, which defaults to the .plog folder in the working directory. ramdisk_path controls the location of states for Volatile<T>, which defaults to tmpfs (ramdisk). reset controls weather to clean up the persisted state when a Derecho service shuts down. We default this to true. Please set reset to false for normal use of Persistent<T>.
We also allow applications to specify configuration options on the command line. Any command line configuration options override the equivalent option in configuration file. To use this feature while still accepting application-specific command-line arguments, we suggest using the following code:
#define NUM_OF_APP_ARGS () // specify the number of application arguments.
int main(int argc, char* argv[]) {
if((argc < (NUM_OF_APP_ARGS+1)) ||
((argc > (NUM_OF_APP_ARGS+1)) && strcmp("--", argv[argc - NUM_OF_APP_ARGS - 1]))) {
cout << "Invalid command line arguments." << endl;
cout << "USAGE:" << argv[0] << "[ derecho-config-list -- ] application-argument-list" << endl;
return -1;
}
Conf::initialize(argc, argv); // pick up configurations in the command line list
// pick up the application argument list and continue ...
...
}Then, call the application as follows, assuming the application's name is app:
$ app --DERECHO/local_id=0 --PERS/reset=false -- <application-argument-list>Please refer to the bandwidth_test application for more details.
There are some sample programs in the folder applications/demos that can be run to test the installation. In addition, there are some performance tests in the folder applications/tests/performance_tests that you may want to use to measure the performance Derecho achieves on your system. To be able to run the tests, you need a minimum of two machines connected by RDMA. The RDMA devices on the machines should be active. In addition, you need to run the following commands to install and load the required kernel modules for using RDMA hardware:
sudo apt-get install rdmacm-utils ibutils libmlx4-1 infiniband-diags libmthca-dev opensm ibverbs-utils libibverbs1 libibcm1 libibcommon1
sudo modprobe -a rdma_cm ib_uverbs ib_umad ib_ipoib mlx4_ib iw_cxgb3 iw_cxgb4 iw_nes iw_c2 ib_mthca
Depending on your system, some of the modules might not load which is fine.
RDMA requires memory pinning of memory regions shared with other nodes. There's a limit on the maximum amount of memory a process can pin, typically 64 KB, which Derecho easily exceeds. Therefore, you need to set this to unlimited. To do so, append the following lines to /etc/security/limits.conf:
* [username] hard memlock unlimited
* [username] soft memlock unlimited
Derecho maintains many TCP connections as well as disk files for large scale setups. We recommend raise the limit of maximum number of open files by appending the following lines to /etc/security/limits.conf:
* hard nofile 10240
* soft nofile 10240
where [username] is your linux username. A * in place of the username will set this limit to unlimited for all users. Log out and back in again for the limits to reapply. You can test this by verifying that ulimit -l outputs unlimited in bash.
The persistence layer of Derecho stores durable logs of updates in memory-mapped files. Linux also limits the size of memory-mapped files to a small size that Derecho usually exceeds, so you will need to set the system parameter vm.overcommit_memory to 1 for persistence to work. To do this, run the command
sysctl -w vm.overcommit_memory=1
A simple test to see if your setup is working is to run the test bandwidth_test from applications/tests/performance_tests. To run it, go to two of your machines (nodes), cd to Release/src/applications/tests/performance_tests and run ./bandwidth_test 2 0 100000 0 on both. As a confirmation that the experiment finished successfully, the first node will write a log of the result in the file data_derecho_bw, which will be something along the lines of 2 0 10240 300 100000 0 5.07607. Full experiment details including explanation of the arguments, results and methodology is explained in the source documentation for this program.
The file simple_replicated_objects.cpp within applications/demos shows a complete working example of a program that sets up and uses a Derecho group with several Replicated Objects. You can read through that file if you prefer to learn by example, or read on for an explanation of how to use various features of Derecho.
One of the core building blocks of Derecho is the concept of a Replicated Object. This provides a simple way for you to define state that is replicated among several machines and a set of RPC functions that operate on that state.
A Replicated Object is any class that (1) is serializable with the mutils-serialization framework and (2) implements a static method called register_functions().
The mutils-serialization library should have more documentation on making objects serializable, but the most straightforward way is to inherit mutils::ByteRepresentable, use the macro DEFAULT_SERIALIZATION_SUPPORT, and write an element-by-element constructor. The register_functions() method is how your class specifies to Derecho which of its methods should be converted to RPC functions and what their numeric "function tags" should be. It should return a std::tuple containing a pointer to each RPC-callable method, wrapped in the template functions derecho::rpc::tag_p2p or derecho::rpc::tag_ordered, depending on how the method will be invoked. Methods wrapped in tag_p2p can be called by peer-to-peer RPC messages and must not modify the state of the Replicated Object, while methods wrapped in tag_ordered can be called by ordered-multicast RPC messages and can modify the state of the object. We have provided a default implementation of this function that is generated with the macros REGISTER_RPC_FUNCTIONS, P2P_TARGETS, and ORDERED_TARGETS; the P2P_TARGETS and ORDERED_TARGETS macros tag their arguments as P2P and ordered-callable RPC methods, respectively, while the REGISTER_RPC_FUNCTIONS macro combines the results of these macros. Here is an example of a Replicated Object declaration that uses the default implementation macros:
class Cache : public mutils::ByteRepresentable {
std::map<std::string, std::string> cache_map;
public:
void put(const std::string& key, const std::string& value);
std::string get(const std::string& key) const;
bool contains(const std::string& key) const;
bool invalidate(const std::string& key);
Cache() : cache_map() {}
Cache(const std::map<std::string, std::string>& cache_map) : cache_map(cache_map) {}
DEFAULT_SERIALIZATION_SUPPORT(Cache, cache_map);
REGISTER_RPC_FUNCTIONS(Cache,
ORDERED_TARGETS(put, invalidate),
P2P_TARGETS(get, contains));
};This object has one field, cache_map, so the DEFAULT_SERIALIZATION_SUPPORT macro is called with the name of the class and the name of this field. The second constructor, which initializes the field from a parameter of the same type, is required for serialization support. The object has two read-only RPC methods that should be invoked by peer-to-peer messages, get and contains, so these method names are passed to the P2P_TARGETS macro; similarly, it has two read-write RPC methods that should be invoked by ordered multicasts, put and invalidate, so these method names are passed to the ORDERED_TARGETS macro. The numeric function tags generated by REGISTER_RPC_FUNCTIONS can be re-generated with the macro RPC_NAME, so these functions can later be called by using the tags RPC_NAME(put), RPC_NAME(get) RPC_NAME(contains), and RPC_NAME(invalidate).
Derecho organizes nodes (machines or processes in a system) into Groups, which can then be divided into subgroups and shards. Any member of a Group can communicate with any other member, and all run the same group-management service that handles failures and accepts new members. Subgroups, which are any subset of the nodes in a Group, correspond to Replicated Objects; each subgroup replicates the state of a Replicated Object and any member of the subgroup can handle RPC calls on that object. Shards are disjoint subsets of a subgroup that each maintain their own state, so one subgroup can replicate multiple instances of the same type of Replicated Object. A Group must be statically configured with the types of Replicated Objects it can support, but the number of subgroups and their exact membership can change at runtime according to functions that you provide.
Note that more than one subgroup can use the same type of Replicated Object, so there can be multiple independent instances of a Replicated Object in a Group even if those subgroups are not sharded. A subgroup is usually identified by the type of Replicated Object it implements and an integral index number specifying which subgroup of that type it is.
To start using Derecho, a process must either start or join a Group by constructing an instance of derecho::Group, which then provides the interface for interacting with other nodes in the Group. (The process will start a new Group if it is configured as the Group leader, otherwise it joins the existing group by contacting the configured leader). A derecho::Group expects a set of variadic template parameters representing the types of Replicated Objects that it can support in its subgroups. For example, this declaration is a pointer to a Group object that can have subgroups of type LoadBalancer, Cache, and Storage:
std::unique_ptr<derecho::Group<LoadBalancer, Cache, Storage>> group;In order to start or join a Group, all members (including processes that join later) must define a function that provides the membership (as a subset of the current View) for each subgroup. The membership function's input is the list of Replicated Object types, the current View, and the previous View if there was one. Its return type is a std::map mapping each Replicated Object type to a vector representing all the subgroups of that type (since there can be more than one subgroup that implements the same Replicated Object type). Each entry in this vector is another vector, whose size indicates the number of shards the subgroup should be divided into, and whose entries are SubViews describing the membership of each shard. For example, if the membership function's return value is named members, then members[std::type_index(typeid(Cache))][0][2] is a SubView identifying the members of the third shard of the first subgroup of type "Cache."
The default membership function: Derecho provides a default subgroup membership function that automatically assigns nodes from the Group into disjoint subgroups and shards, given a policy (an instance of SubgroupAllocationPolicy or ShardAllocationPolicy) that describes the desired number of nodes in each subgroup/shard. The function assigns nodes in ascending rank order, and leaves any "extra" nodes (not needed to fully populate all subgroups) at the end (highest rank) of the membership list. During a View change, it attempts to preserve the correct number of nodes in each shard without re-assigning any nodes to new roles. It does this by copying the subgroup membership from the previous View as much as possible, and assigning idle nodes from the end of the Group's membership list to replace failed members of subgroups.
There are several helper functions in subgroup_functions.hpp that construct SubgroupAllocationPolicy objects for different scenarios, to make it easier to set up the default subgroup membership function. Here is an example of how the default membership function could be configured for two types of Replicated Objects using these functions:
derecho::SubgroupInfo subgroup_function {derecho::DefaultSubgroupAllocator({
{std::type_index(typeid(Foo)), derecho::one_subgroup_policy(derecho::even_sharding_policy(2,3))},
{std::type_index(typeid(Bar)), derecho::identical_subgroups_policy(
2, derecho::even_sharding_policy(1,3))}
})};Based on the policies constructed for the constructor argument of DefaultSubgroupAllocator, the function will create one subgroup of type Foo, with two shards of 3 members each. Next, it will create two subgroups of type Bar, each of which has only one shard of size 3. Note that the order in which subgroups are allocated is the order in which their Replicated Object types are listed in the Group's template parameters, so this instance of the default subgroup allocator will assign the first 6 nodes to the Foo subgroup and the second 6 nodes to the Bar subgroups the first time it runs.
The subgroup membership function's SubgroupAllocationPolicy objects can also be constructed from JSON strings, which allows you to change the allocation policy without recompiling code. To use this feature, add the option json_layout_path to the Derecho config file, specifying the (relative or absolute) path to a JSON file, and construct the DefaultSubgroupAllocator using the templated make_subgroup_allocator function. The template parameters to this function must be the same Replicated Objects as the template parameters for the Group object, in the same order; for example, this subgroup function would be used for derecho::Group<Foo, Bar>:
derecho::SubgroupInfo subgroup_function{derecho::make_subgroup_allocator<Foo, Bar>()};The JSON file specified with the json_layout_path option should contain an array with one entry per Replicated Object type, in the same order as these types are listed in the template parameters. Each array entry should be an (JSON) object with a property named "layout", whose value is an array of one or more objects describing allocation policies for each subgroup of this type. An allocation policy object in JSON has fields very similar to the fields of ShardAllocationPolicy object: min_nodes_by_shard, max_nodes_by_shard, delivery_modes_by_shard, and profiles_by_shard. Unlike a ShardAllocationPolicy object, however, identical_subgroups and even_shards cannot be used; each of these fields must be an array of length equal to the number of shards desired. Here is an example of a JSON file that creates the same allocation policies as the arguments to the DefaultSubgroupAllocator constructor above:
[
{
"type_alias": "Foo",
"layout": [
{
"min_nodes_by_shard": [2, 2],
"max_nodes_by_shard": [3, 3],
"delivery_modes_by_shard": ["Ordered", "Ordered"],
"profiles_by_shard": ["DEFAULT", "DEFAULT"]
}
]
},
{
"type_alias": "Bar",
"layout": [
{
"min_nodes_by_shard": [1],
"max_nodes_by_shard": [3],
"delivery_modes_by_shard": ["Ordered"],
"profiles_by_shard": ["DEFAULT"]
},
{
"min_nodes_by_shard": [1],
"max_nodes_by_shard": [3],
"delivery_modes_by_shard": ["Ordered"],
"profiles_by_shard": ["DEFAULT"]
}
]
}
]
The field "type_alias" is optional for Derecho (it is only used by Cascade), but it is useful to help keep track of which array entry corresponds to which subgroup type. The layout policies will actually be applied to Replicated Object types in the order they appear in the template parameters for Group and make_subgroup_allocator, regardless of what name is used in "type_alias."
Instead of using a separate JSON file, the same JSON layout policy can be embedded directly in the Derecho config file using the option json_layout, whose value will be interpreted as a JSON string. It is an error to use both json_layout and json_layout_path in the same config file.
Reserved node IDs: The default subgroup allocation function's behavior can be significantly changed by using the optional field reserved_node_ids_by_shard in ShardAllocationPolicy, which has an equivalent optional field "reserved_node_ids_by_shard" in the JSON layout syntax. This field specifies a set of node IDs that should always be assigned to each shard (if they exist in the current View), regardless of where those nodes appear in the rank order. If shards have reserved node IDs, the allocation function will always assign those node IDs to the shards that reserved them, and then assign any remaining nodes in the default fashion (round robin in ascending rank order). If multiple shards from different subgroups reserve the same node IDs, those nodes will be assigned to all of the shards that reserved them, and thus be members of more than one subgroup. However, multiple shards in the same subgroup cannot reserve the same node ID (this will result in a configuration error), since shards by definition must be disjoint.
Here is an example of a JSON layout string that uses "reserved_node_ids_by_shard" to make the Bar subgroup's (only) shard co-resident with members of both shards of the Foo subgroup:
[
{
"type_alias": "Foo",
"layout": [
{
"min_nodes_by_shard": [2, 2],
"max_nodes_by_shard": [3, 3],
"reserved_nodes_by_shard": [[1, 2, 3], [4, 5, 6]],
"delivery_modes_by_shard": ["Ordered", "Ordered"],
"profiles_by_shard": ["DEFAULT", "DEFAULT"]
}
]
},
{
"type_alias": "Bar",
"layout": [
{
"min_nodes_by_shard": [1],
"max_nodes_by_shard": [3],
"reserved_nodes_by_shard": [[3, 4, 5]],
"delivery_modes_by_shard": ["Ordered"],
"profiles_by_shard": ["DEFAULT"]
}
]
}
]
Defining a custom membership function: If the default membership function's node-allocation algorithm doesn't fit your needs, you can define own subgroup membership function. The demo program overlapping_replicated_objects.cpp shows a relatively simple example of a user-defined membership function. In this program, the SubgroupInfo contains a C++ lambda function that implements the shard_view_generator_t type signature and handles subgroup assignment for Replicated Objects of type Foo, Bar, and Cache:
[](const std::vector<std::type_index>& subgroup_type_order,
const std::unique_ptr<derecho::View>& prev_view, derecho::View& curr_view) {
derecho::subgroup_allocation_map_t subgroup_allocation;
for(const auto& subgroup_type : subgroup_type_order) {
derecho::subgroup_shard_layout_t subgroup_layout(1);
if(subgroup_type == std::type_index(typeid(Foo)) || subgroup_type == std::type_index(typeid(Bar))) {
// must have at least 3 nodes in the top-level group
if(curr_view.num_members < 3) {
throw derecho::subgroup_provisioning_exception();
}
std::vector<derecho::node_id_t> first_3_nodes(&curr_view.members[0], &curr_view.members[0] + 3);
//Put the desired SubView at subgroup_layout[0][0] since there's one subgroup with one shard
subgroup_layout[0].emplace_back(curr_view.make_subview(first_3_nodes));
//Advance next_unassigned_rank by 3, unless it was already beyond 3, since we assigned the first 3 nodes
curr_view.next_unassigned_rank = std::max(curr_view.next_unassigned_rank, 3);
} else { //subgroup_type == std::type_index(typeid(Cache))
// must have at least 6 nodes in the top-level group
if(curr_view.num_members < 6) {
throw derecho::subgroup_provisioning_exception();
}
std::vector<derecho::node_id_t> next_3_nodes(&curr_view.members[3], &curr_view.members[3] + 3);
subgroup_layout[0].emplace_back(curr_view.make_subview(next_3_nodes));
curr_view.next_unassigned_rank += 3;
}
subgroup_allocation.emplace(subgroup_type, std::move(subgroup_layout));
}
return subgroup_allocation;
};For all three types of Replicated Object, the function creates one subgroup and one shard. For the Foo and Bar subgroups, it assigns first three nodes in the current View's members list (thus, these subgroups are co-resident on the same three nodes), while for the Cache subgroup it assigns nodes 3 to 6 on the current View's members list. Note that if there are not enough members in the current view to assign 3 nodes to each subgroup, the function throws derecho::subgroup_provisioning_exception. This is how subgroup membership functions indicate to the view management logic that a view has suffered too many failures to continue executing (it is "inadequately provisioned") and must wait for more members to join before accepting any more state updates.
Although the subgroup allocation function is the most important part of constructing a derecho::Group, it requires a few additional parameters.
- A set of callback functions that will be notified when each Derecho message is delivered to the node (stability callback) or persisted to disk (persistence callback). These can be null, and are probably not useful if you're only using the Replicated Object features of Derecho (since the "messages" will be serialized RPC calls).
- A set of View upcalls that will be notified when the group experiences a View change event (nodes fail or join the group). This is optional and can be empty, but it can be useful for adding additional failure-handling or load-balancing behavior to your application.
- For each template parameter in the type of
derecho::Group, its constructor will expect an additional argument of typederecho::Factory, which is a function or functor that constructs instances of the Replicated Object given parameters of typePersistentRegistry*andsubgroup_id_t(thePersistentRegistrypointer is used to initializePersistent<T>fields, while thesubgroup_id_tparameter tells the object which subgroup it has been assigned to).
Once a process has joined a Group and one or more subgroups, it can invoke RPC functions on any of the Replicated Objects in the Group. The options a process has for invoking RPC functions depend on its membership status:
- A node can perform an ordered send to invoke an RPC function on a Replicated Object only when it is a member of that object's subgroup and shard. This sends a multicast to all other members of the object's shard, and guarantees that the multicast will be delivered in order (so the function call will take effect at the same time on every node). An ordered send waits for responses from each member of the shard (specifically, it provides a set of Future objects that can be used to wait for responses), unless the function invoked had a return type of
void, in which case there are no responses to wait for. - A node can perform a P2P send to invoke a read-only RPC function on any Replicated Object, regardless of whether it is a member of that object's subgroup and shard. These peer-to-peer operations send a message directly to a specific node, and it is up to the sender to pick a node that is a member of the desired object's shard. They cannot be used for mutative RPC function calls because they do not guarantee ordering of the message with respect to any other (peer-to-peer or "ordered") message, and could only update one replica at a time.
Ordered sends are invoked through the Replicated interface, whose template parameter is the type of the Replicated Object it communicates with. You can obtain a Replicated by using Group's get_subgroup method, which uses a template parameter to specify the type of the Replicated Object and an integer argument to specify which subgroup of that type (remember that more than one subgroup can implement the same type of Replicated Object). For example, this code retrieves the Replicated object corresponding to the second subgroup of type Cache:
Replicated<Cache>& cache_rpc_handle = group->get_subgroup<Cache>(1);The ordered_send method uses its template parameter, which is an integral "function tag," to specify which RPC function it will invoke; if you are using the REGISTER_RPC_FUNCTIONS macro, the function tag will be the integer generated by the RPC_NAME macro applied to the name of the function. Its arguments are the arguments that will be passed to the RPC function call, and it returns an instance of derecho::rpc::QueryResults with a template parameter equal to the return type of the RPC function. Using the Cache example from earlier, this is what RPC calls to the "put" and "contains" functions would look like:
cache_rpc_handle.ordered_send<RPC_NAME(put)>("Foo", "Bar");
derecho::rpc::QueryResults<bool> results = cache_rpc_handle.ordered_send<RPC_NAME(contains)>("Foo");P2P (peer-to-peer) sends are invoked through the PeerCaller interface, which is exactly like the Replicated interface except that it only provides the p2p_send function. PeerCaller objects are provided through the get_nonmember_subgroup method of Group, which works exactly like get_subgroup (except for the assumption that the caller is not a member of the requested subgroup). For example, this is how a process that is not a member of the second Cache-type subgroup would get a PeerCaller to that subgroup:
PeerCaller<Cache>& p2p_cache_handle = group->get_nonmember_subgroup<Cache>(1);When invoking a P2P send, the caller must specify, as the first argument, the ID of the node to communicate with. The caller must ensure that this node is actually a member of the subgroup that the PeerCaller targets (though it can be in any shard of that subgroup). Nodes can find out the current membership of a subgroup by calling the get_subgroup_members method on the Group, which uses the same template parameter and argument as get_subgroup to select a subgroup by type and index. For example, assuming Cache subgroups are not sharded, this is how a non-member process could make a call to get, targeting the first node in the second subgroup of type Cache:
std::vector<derecho::node_id_t> cache_members = group.get_subgroup_members<Cache>(1)[0];
derecho::rpc::QueryResults<std::string> results = p2p_cache_handle.p2p_send<RPC_NAME(get)>(cache_members[0], "Foo");The result of an ordered send is a slightly complex object, because it must contain a std::future for each member of the subgroup, but the membership of the subgroup might change during the query invocation. Thus, a QueryResults object is actually itself a future, which is fulfilled with a map from node IDs to futures as soon as Derecho can guarantee that the query will be delivered in a particular View. (The node IDs in the map are the members of the subgroup in that View). Each std::future in the map will be fulfilled with either the response from that node or a node_removed_from_group_exception, if a View change occurred after the query was delivered but before that node had a chance to respond.
By the time the caller sees this exception, the view will have been updated. Thus a caller that wishes to reissue a request could do so immediately after the exception is caught: it can already look up the new membership, select a new target, and send a new request. On the other hand, notice that Derecho provides no indication of whether the target that failed did so before or after the original request was received. Thus if your target might have taken some action (like issuing an update request), you may have to include application-layer logic to make sure your reissued request won't be performed twice if the initial request actually got past the update step, and the failure occurred later. A simple way to do this is to make your requests idempotent, for example by including a request-id and an "this is a retry" flag, and if the flag is true, having the group member check to see if that request-id has already been performed.
As an example, this code waits for the responses from each node and combines them to ensure that all replicas agree on an item's presence in the cache:
derecho::rpc::QueryResults<bool> results = cache_rpc_handle.ordered_send<RPC_NAME(contains)>("Stuff");
bool contains_accum = true;
for(auto& reply_pair : results.get()) {
bool contains_result = reply_pair.second.get();
contains_accum = contains_accum && contains_result;
}Note that the type of reply_pair is std::pair<derecho::node_id_t, std::future<bool>>, which is why a node's response is accessed by writing reply_pair.second.get().
Derecho allows tracking data update history with a version vector in memory or persistent storage. A new class template is introduced for this purpose: Persistent<T,ST>. In a Persistent instance, data is managed in an in-memory object of type T (we call it the "current object") along with a log in a datastore specified by storage type ST. The log can be indexed using a version number, an index, or a timestamp. A version number is a 64-bit integer attached to each version; it is managed by the Derecho SST and guaranteed to be monotonic. A log is also an array of versions accessible using zero-based indices. Each log entry also has an attached timestamp (microseconds) indicating when this update happened according to the local real-time clock. To enable this feature, we need to manage the data in a serializable object T, and define a member of type Persistent<T> in the Replicated Object in a relevant group. Persistent_typed_subgroup_test.cpp gives an example.
/**
* Example for replicated object with Persistent<T>
*/
class PFoo : public mutils::ByteRepresentable {
Persistent<int> pint;
public:
virtual ~PFoo() noexcept (true) {}
int read_state() const {
return *pint;
}
bool change_state(int new_int) {
if(new_int == *pint) {
return false;
}
*pint = new_int;
return true;
}
// constructor with PersistentRegistry
PFoo(PersistentRegistry * pr) : pint(nullptr,pr) {}
PFoo(Persistent<int> & init_pint) : pint(std::move(init_pint)) {}
DEFAULT_SERIALIZATION_SUPPORT(PFoo, pint);
REGISTER_RPC_FUNCTIONS(PFoo, P2P_TARGETS(read_state), ORDERED_TARGETS(change_state));
};For simplicity, the versioned type is int in this example. You set it up in the same way as a non-versioned member of a replicated object, except that you need to pass the PersistentRegistry from the constructor of the replicated object to the constructor of the Persistent<T>. Derecho uses PersistentRegistry to keep track of all the Persistent<T> objects in a single Replicated Object so that it can create versions on updates. The Persistent<T> constructor registers itself in the registry.
By default, the Persistent<T> stores its log in the file-system (in a folder called .plog in the current directory). Applications can specify memory as the storage location by setting the second template parameter: Persistent<T,ST_MEM> (or Volatile<T> as syntactic sugar). We are working on more storage types including NVM.
Once the version vector is set up with Derecho, the application can query the value with the get() APIs in Persistent<T>. In persistent_temporal_query_test.cpp, a temporal query example is illustrated.
We are committed to supporting Derecho with RDMA on 1000 (or even more) physical nodes, one application instance per node. On a machine that actually allows some small number K of applications to share an RDMA NIC, we would even be happy to help get things working with k*1000's of group members... eventually. However, we do not recommend that Derecho developers start by trying to work at that scale before gaining experience at smaller scales. Even launching a Derecho test program at that scale would be very challenging, and we will only be able to help if the team undertaking this has a good level of experience with the system at smaller scales.
Moreover, "running Derecho" is a bit of a broad term. For example, we view it as highly experimental to run with TCP on more than 16-32 nodes (we do want to scale over TCP, but it will take time). So we would not recommend attempting to run Derecho on TCP at 1000-node scale, no matter how good your reasons: this very likely will be hard to engineer into our TCP layering, which actually runs over LibFabric, and would likely expose LibFabric scaling and performance issues. Similarly, it is not wise to try and run 1000 Derecho application instances on, for example, 2 AWS servers, using AWS container virtualization (or the same comment with Azure, or Google, or IBM, or whatever as your provider). That will never work, due to timeouts, and we will not try to support that sort of thing: it would be a waste of our time. Container virtualization isn't capable of supporting this kind of application.
Additionally, it is important for you as the developer to realize that launching on 1000 physical nodes is hard, and you will spend days or weeks before this is fully stable. Once Derecho is up and running, you depend only on our layers and those are designed to work at large scale. But Linux is not designed to boot an application with 1000 members all of which make connections to one-another (so you get 1,000,000 connections right as it starts), plus the file system issues mentioned above. Linux isn't normally exposed to this sort of nearly simultaneous burst of load. Thus, Linux can be overwhelmed.
Even on HPC systems, which can support MPI at that scale, because MPI doesn't use an all-to-all connection pattern, we have seen these kinds of difficulties at massive scale. In MPI there is one leader and N-1 followers, so the primary pattern that arises is really 1-N connections (more accurately, they do have some cases at runtime (like AllReduce) that can create K x K patterns. I don't know how much success those folks have had with K>=1000, though. My impression is that All Reduce normally runs on a significantly smaller scale. Other KxK situations on MPI are probably delicate to initialize, too.)
At Cornell, up to now, our largest experiments involved cases where we benchmarked RDMC (not the full Derecho) on 1000's of nodes at the LLNL supercomputer center. And it was a nightmare getting to the point where that worked. In the end, we actually had a special batch script to launch them 50 at a time, and have them connect in batches, to avoid overloading the file system and TCP layer.
Our largest Derecho experiments have been on a Texas supercomputer, where we had successful and completely stable runs on 256 physical nodes and probably could have pushed towards 1024 or more had we not run out of credits: "renting" 1000's of non-virtualized nodes is expensive. Then just as we applied for more credit, they decommissioned the entire machine (Stampede-1). So that whole line of experiments ended abruptly. Still, we do think it could have been carried quite a bit further. In this mode, we felt we were experimenting on a use case and deployment of a kind that Derecho needs to support.
So... the concept of Derecho at 1000's of nodes is something we definitely intend to support, in some specific situations where the goal makes sense, the underlying infrastructure should be able to do it, and where we have access to do debugging of our own RDMA layers during the startup. But this isn't a sensible thing to do as your very first Derecho deployment!