Briefly, this error occurs when Elasticsearch attempts to write new repository data while a previous write operation is still in progress. This could be due to concurrent snapshot operations or a network issue. To resolve this, you can try the following: 1) Ensure only one snapshot operation is running at a time. 2) Check network connectivity and stability. 3) If the error persists, consider deleting and recreating the repository, but ensure you have a backup of your data before doing so.
This guide will help you check for common problems that cause the log ” Trying to write new repository data over unfinished write; repo [{}] is at ” to appear. To understand the issues related to this log, read the explanation below about the following Elasticsearch concepts: repositories, blobstore, repository.
Overview
An Elasticsearch snapshot provides a backup mechanism that takes the current state and data in the cluster and saves it to a repository (read snapshot for more information). The backup process requires a repository to be created first. The repository needs to be registered using the _snapshot endpoint, and multiple repositories can be created per cluster. The following repository types are supported:
Repository types
| Repository type | Configuration type |
|---|---|
| Shared file system | Type: “fs” |
| S3 | Type : “s3” |
| HDFS | Type :“hdfs” |
| Azure | Type: “azure” |
| Google Cloud Storage | Type : “gcs” |
Examples
To register an “fs” repository:
PUT _snapshot/my_repo_01
{
"type": "fs",
"settings": {
"location": "/mnt/my_repo_dir"
}
}Notes and good things to know
- S3, HDFS, Azure and Google Cloud require a relevant plugin to be installed before it can be used for a snapshot.
- The setting, path.repo: /mnt/my_repo_dir needs to be added to elasticsearch.yml on all the nodes if you are planning to use the repo type of file system. Otherwise, it will fail.
- When using remote repositories, the network bandwidth and repository storage throughput should be high enough to complete the snapshot operations normally, otherwise you will end up with partial snapshots.
Log Context
Log “Trying to write new repository data over unfinished write; repo [{}] is at ” classname is BlobStoreRepository.java.
We extracted the following from Elasticsearch source code for those seeking an in-depth context :
final RepositoryMetadata meta = getRepoMetadata(currentState);
final String repoName = metadata.name();
final long genInState = meta.generation();
final boolean uninitializedMeta = meta.generation() == RepositoryData.UNKNOWN_REPO_GEN || bestEffortConsistency;
if (uninitializedMeta == false && meta.pendingGeneration() != genInState) {
logger.info("Trying to write new repository data over unfinished write; repo [{}] is at " +
"safe generation [{}] and pending generation [{}]"; meta.name(); genInState; meta.pendingGeneration());
}
assert expectedGen == RepositoryData.EMPTY_REPO_GEN || uninitializedMeta
|| expectedGen == meta.generation() :
"Expected non-empty generation [" + expectedGen + "] does not match generation tracked in [" + meta + "]";
