Publicação
Causal Consistency Verification in Restful Systems
| Resumo: | Replicated systems cannot maintain both availability and (strong) consistency when exposed to network partitions. Strong consistency requires every read to return the last written value, which can lead clients to experience high latency or even timeout errors. Replicated applications usually rely on weak consistency, since clients can perform operations contacting a single replica, leading to decreased latency and increased availability. Causal consistency is a weak consistency model, however, it is the strongest one for highly available systems. Many applications are switching to this particular consistency model, since it ensures users never observe data items before they observe the ones that influenced their creation. Verifying if applications satisfy the consistency they claim to provide is no easy task. In this dissertation, we propose an algorithm to verify causal consistency in RESTful applications. Our approach adopts a black box testing, where multiple concurrent clients execute operations in a service and records the log of interactions. This log of interactions is then processed to verify if the results respect causal consistency. The key challenge is to infer causal dependencies among operations executed in different clients without adding any additional metadata to the data maintained by the service. When considering a particular operation, the algorithm builds a new dependency graph that considers one of the possible justifications the operation might have, but if this justification results in failure further ahead in the processing, it is necessary to build another graph considering another justification of that same operation. The algorithm relies on recursion in order to achieve this backtracking behaviour. If the algorithm is able to build a graph containing every operation present in the log, where the chosen justifications remain valid until the end of the processing, it outputs that the execution corresponding to that log satisfies causal consistency. The evaluation confirms that the algorithm is able to detect violations when feeding either small or large logs representing executions of RESTful applications that do not satisfy causal consistency. |
|---|---|
| Autores principais: | Rodrigues, Hugo Miguel Grilo |
| Assunto: | Distributed Systems RESTful Applications Causal Consistency VectorClocks Jepsen JepREST |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Replicated systems cannot maintain both availability and (strong) consistency when exposed to network partitions. Strong consistency requires every read to return the last written value, which can lead clients to experience high latency or even timeout errors. Replicated applications usually rely on weak consistency, since clients can perform operations contacting a single replica, leading to decreased latency and increased availability. Causal consistency is a weak consistency model, however, it is the strongest one for highly available systems. Many applications are switching to this particular consistency model, since it ensures users never observe data items before they observe the ones that influenced their creation. Verifying if applications satisfy the consistency they claim to provide is no easy task. In this dissertation, we propose an algorithm to verify causal consistency in RESTful applications. Our approach adopts a black box testing, where multiple concurrent clients execute operations in a service and records the log of interactions. This log of interactions is then processed to verify if the results respect causal consistency. The key challenge is to infer causal dependencies among operations executed in different clients without adding any additional metadata to the data maintained by the service. When considering a particular operation, the algorithm builds a new dependency graph that considers one of the possible justifications the operation might have, but if this justification results in failure further ahead in the processing, it is necessary to build another graph considering another justification of that same operation. The algorithm relies on recursion in order to achieve this backtracking behaviour. If the algorithm is able to build a graph containing every operation present in the log, where the chosen justifications remain valid until the end of the processing, it outputs that the execution corresponding to that log satisfies causal consistency. The evaluation confirms that the algorithm is able to detect violations when feeding either small or large logs representing executions of RESTful applications that do not satisfy causal consistency. |
|---|