CLOUDFLARE, INC.

A compute server receives a request that triggers execution of a code piece out of multiple code pieces. A single process at the compute server executes the code piece, which is run in an isolated execution environment. Each other code piece runs in other isolated execution environments respectively and executed by the single process. The code piece, when executed, modifies a response to the request. The response is generated based at least in part on the executed code piece. The generated response is transmitted.

Purging resources from a cache in a distributed networked system is described. A first data center of the distributed networked system receives a purge request to purge a resource from cache. If the purge request does not include a cache key, the first data center determines whether the purge request is valid, and if valid, purges the resource from cache of the first data center, generates a cache key for the resource, and causes the purge request that includes the generated cache key to be sent to other data centers of the distributed networked system for purging the resource from cache. If the purge request includes a cache key, the first data center skips determining whether the purge request is valid and purges the resource from cache based on the cache key.

A tunnel is established between a first edge server of a distributed edge compute and routing service and a tunnel client residing on an origin server. Routing rules are installed in the edge servers of the distributed edge compute and routing service to reach the first edge server. The routing rules are based at least in part on traffic information gathered from processing other traffic that traverses the distributed edge compute and routing service. A request for content served by the origin server through the tunnel is received at a second edge server of the distributed edge compute and routing service. A path from the second edge server to the first edge server is determined based on the routing rules. The request is transmitted on the determined path. The first edge server receives the request and transmits the request to the origin server over the tunnel.

A first compute server of a distributed cloud computing network receives a request from a first client device for an object to be handled by an object worker that includes a single instantiation of a piece of code that solely controls reading and writing access to the first object. A determination is made that the object worker is instantiated for the object and is currently running in the first compute server, and the piece of code processes the first request. The first compute server receives a message to be processed by the first object worker from a second compute server. The message includes a second request for the object from a second client device connected to the second compute server. The piece of code processes the message and transmits a reply to the second compute server.

Methods and system of traffic load balancing between a plurality of Points of Presence (PoP) of a cloud computing infrastructure are described. A first PoP of multiple PoPs of cloud computing infrastructure that provides a cloud computing service receives a packet. The packet includes as a destination address an anycast address advertised by the first PoP for reaching the cloud computing service. The first PoP identifies a network address of a second PoP that is different from the first PoP. The first PoP forwards the packets as an encapsulated packet to the second PoP to be processed in the second PoP according to the cloud computing service.

A server receives internet traffic from a client device. The server is one of multiple servers of a distributed cloud computing network which are each associated with a set of server identity(ies) including a server/data center certification identity. The server processes, at layer 3, the internet traffic including participating in a layer 3 DDoS protection service. If the traffic is not dropped by the layer 3 DDoS protection service, further processing is performed. The server determines whether it is permitted to process the traffic at layers 5-7 including whether it is associated with a server/data center certification identity that meets a selected criteria for the destination of the internet traffic. If the server does not meet the criteria, it transmits the traffic to another one of the multiple servers for processing the traffic at layers 5-7.

A request from a client device is received at a first one of a plurality of compute nodes at a first one of a plurality of data centers of a distributed cloud computing network. A destination of the request is determined. An optimized route for transmitting the request toward an origin server that corresponds with the destination of the request is determined, where the optimized route is based on at least in part on probe data between data centers of the distributed cloud computing network for a plurality of transit connections, and where the optimized route has an IP address that encodes an identification of which of the plurality of transit connections is to be used to deliver the request. The request is transmitted to a next hop as defined by the optimized route over the identified one of the plurality of transit connections.

A GRE tunnel is configured between multiple computing devices of a distributed cloud computing network and a single origin router of the origin network. The GRE tunnel has a first GRE endpoint that has an IP address that is shared among the computing devices of the distribute cloud computing network and a second GRE endpoint that has a publicly routable IP address of the origin router. A first computing device receives an IP packet from a client that is destined to an origin server. The first computing device processes the received IP packet and encapsulates the IP packet inside an outer packet to generate a GRE encapsulated packet whose source address is the first GRE endpoint and the destination address is the second GRE endpoint. The GRE encapsulated packet is transmitted over the GRE tunnel to the single origin router.

Inter-process serving of machine learning features from mapped memory for machine learning models is described. ML features are populated in a data structure that is serialized. State data is stored that indicates that reader process(es) are to read from a first memory mapped data file and not a second memory mapped data file. The serialized bytes are stored in the second memory mapped data file and the state data is updated to indicate that the reader process(es) are to read from the second memory mapped data file. A request is received and parsed to prepare keys from attributes of the request. Based on the state data, the serialized bytes are read from the second memory mapped data file that correspond to the keys. The serialized bytes are deserialized and copied to a data structure available to an inference algorithm.

A map of IP addresses of a distributed cloud computing network to one or more groupings is stored. The IP addresses are anycast IP addresses for which compute servers of the distributed cloud computing network share. These IP addresses are to be used as source IP addresses when transmitting traffic to destinations external to the cloud computing network. The map is made available to external destinations. Traffic is received at the distributed cloud computing network that is destined to an external destination. An IP address is selected based on the characteristic(s) applicable for the traffic and the map. The distributed cloud computing network transmits the traffic to the external destination using the selected IP address.