Troubleshooting Amazon CloudSearch

The following topics describe solutions to problems you might encounter when using Amazon CloudSearch.

Uploading Documents

If your document data is not formatted correctly or contains invalid values, you will get errors when you attempt to upload it or use it to configure fields for your domain. Here are some common problems and their solutions:

Deleting All Documents in an Amazon CloudSearch Domain

Amazon CloudSearch currently does not provide a mechanism for deleting all of the documents in a domain.

Amazon CloudSearch Domain Not Scaling Down After Deleting Documents

If your domain has scaled up to accommodate your index size and you delete a large number of documents, the domain scales down the next time the full index is rebuilt. Although the index is automatically rebuilt periodically, to scale down as quickly as possible you can explicitly run indexing when you are done deleting documents.

Document Update Latency

Sending a large volume of single-document batches can increase the amount of time it takes each document to become searchable. If you have a large amount of update traffic, you need to batch your updates. We recommend using a batch size close to the 5 MB limit. For more information, see Creating Document Batches.

You can load up to 10,000 document batches per day (every 24 hours), with each batch size up to 5 MB. Loading more data per day significantly increases the latency of document updates. To mitigate this risk, you can increase your update capacity by selecting a larger instance type. For more information, see Configuring Scaling Options in Amazon CloudSearch.

Large Number of 5xx Errors When Uploading Documents to an Amazon CloudSearch Domain

If you parallelize uploads and your domain is on a search.small instance, you might experience an unacceptably high rate of 504 or 507 errors. Setting the desired instance type to a larger instance type will increase your update capacity and reduce the error rate. For more information about handling 5xx errors, see Handling Errors. For information about prescaling your domain to increase upload capacity, see Configuring Scaling Options in Amazon CloudSearch.

Search Latency and Timeouts in Amazon CloudSearch

If you are experiencing slow response times, frequently encountering internal server errors (typically 507 or 509 errors), or seeing the number of instance hours your search domain is consuming increase without a substantial increase in the volume of data you're searching, fine-tuning your search requests to reduce the processing overhead can help. For more information, see Tuning Search Request Performance in Amazon CloudSearch. Increasing the desired replication count can also speed up search request processing. For more information, see Configuring Scaling Options in Amazon CloudSearch.

507 and 509 errors typically indicate that your search service is overloaded. This can be due to the volume or complexity of search requests that you are submitting. Amazon CloudSearch normally scales automatically to handle the load. Because it takes some time to deploy additional search instances, we recommend using an exponential backoff retry policy to temporarily reduce the request rate and minimize request failures. For more information, see Error Retries and Exponential Backoff.

Certain usage patterns, such as submitting complex search queries to a single small search instance, can sometimes result in timeouts without triggering automatic scaling. If you repeatedly experience a high error rate, you can explicitly request additional capacity through the Amazon CloudSearch Service Limit Request form.

Search Latency for Faceted Queries in Amazon CloudSearch

If you are bucketing facet information with the buckets option and experiencing slow query performance, try setting the buckets method to interval. For more information, see Bucketing Facet Information.

Sudden Increase in 5xx Errors When Searching an Amazon CloudSearch Domain

If your search domain experiences a sudden spike in traffic, Amazon CloudSearch responds by adding search instances to your domain to handle the increased load. However, it takes a few minutes to set up the new instances. You are likely to see a temporary increase in 5xx errors until the new instances are ready to start processing requests. For more information about handling 5xx errors, see Handling Errors. For information about pre-scaling your domain to handle an expected spike in search requests, see Configuring Scaling Options in Amazon CloudSearch.

Indexing Failures after Updating Indexing Options in Amazon CloudSearch

If you make changes to a domain's index configuration, in certain cases you might encounter Failed to Validate errors when you run indexing. This means that the index field options you set are not compatible with the documents that are already in your index. Specifically, you changed the type of an index field, and there are documents in your index that contain data that is incompatible with that type. For example, this might happen if you change a literal field to an int field, and some of your documents contain alphanumeric data in that field. When this happens, Amazon CloudSearch sets the status of ALL fields that were being processed to the FailedToValidate state. Rolling back the incompatible configuration change will enable you to successfully rebuild your index. If the change is necessary, you must update or remove the incompatible documents from your index to use the new configuration. If you can't identify the change that caused the error or need assistance identifying the incompatible documents, contact support.

Domain Not Found When Submitting Amazon CloudSearch Requests

You cannot access a 2013-01-01 domain with the 2011-02-01 command line tools or SDKs. Similarly, you cannot access a 2011-02-01 domain with the 2013-01-01 command line tools or SDKs. Make sure you are specifying the correct API version in your request and using the appropriate command line tools or SDK.

Number of Searchable Documents Not Returned with Domain Information

The aws cloudsearch describe-domains and DescribeDomains do not return the number of searchable documents in the domain. To get the number of searchable documents, use the console or submit a matchall request to your domain's search endpoint.

q=matchall&q.parser=structured&size=0

Configuration Service Access Policies Not Working in Amazon CloudSearch

If you have both 2011 and 2013 domains, have configured IAM policies for accessing the configuration service, and are getting not authorized errors, note that the Amazon CloudSearch ARN is different for the 2011-02-01 API and the 2013-01-01 API. To allow users to access both 2011 and 2013 domains, you must allow access to both ARNs in the IAM policy. For example:

{
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "cloudsearch:*",
       ],
      "Resource": "arn:aws:cloudsearch:*",
      "Resource": "arn:aws:cs:*"
    }
  ]
}

If your 2011 policy granted access to particular domains or actions, you must include those restrictions in your policy. Note that the only supported action for 2011 domains is cloudsearch:* and you might encounter other errors when attempting to configure resource-level permissions for domains created with the 2011-01-01 API.

Search and Document Service Access Policies Not Working in Amazon CloudSearch

If you have configured access policies for you domain's search and document service endpoints, but are getting the error 403 Request forbidden by administrative rules, it is likely due to one of the following issues.

For more information, see configure access policies.

Amazon CloudSearch Console Permissions Errors

To access to the console, you must have permissions for the DescribeDomains action. Access to particular domains and actions might be restricted by to the configured IAM access policies. In addition, uploading data from an Amazon S3 bucket or DynamoDB table requires access to those services and resources. For more information about Amazon CloudSearch access policies, see configure access policies.

Using Wildcards to Search Text Fields Doesn't Produce Expected Results

When you submit a search request, the text you're searching for undergoes the same text processing so that it can be matched against the terms that appear in the index. However, no text analysis is performed on the search term when you perform a prefix search. This means that a search for a prefix that ends in s typically won't match the singular version of the term when stemming is enabled. This can happen for any term that ends in s, not just plurals. For example, if you search the actor field in the sample movie data for Anders, there are three matching movies. If you search for Ander*, you get those movies as well as several others. However, if you search for Anders* there are no matches. This is because the term is stored in the index as ander, anders does not appear in the index.

If stemming is preventing your wildcard searches from returning all of the relevant matches, you can suppress stemming for the text field by setting the AlgorithmicStemming option to none, or you can map the data to a literal field instead of a text field.

For more information about how Amazon CloudSearch processes text, see Text Processing in Amazon CloudSearch.

Inconsistent Results When Using Cursors for Deep Paging

When you use a cursor to page through a result set that is sorted by document score (_score), you can get inconsistent results if the index is updated between requests. This can also occur if your domain's replication count is greater than one, because updates are applied in an eventually consistent manner across the instances in the domain. If this is an issue, avoid sorting the results by score. You can either use the sort option to sort by a particular field, or use fq instead of q to specify your search criteria. (Document scores are not calculated for filter queries.)

Certificate Errors When Using an SDK

Because AWS SDKs use the CA certificates from your computer, changes to the certificates on the AWS servers can cause connection failures when you attempt to use an SDK. Error messages vary, but typically contain the following text:

SSL3_GET_SERVER_CERTIFICATE:certificate verify failed

You can prevent these failures by keeping your computer's CA certificates and operating system up-to-date. If you encounter this issue in a corporate environment and do not manage your own computer, you might need to ask an administrator to assist with the update process.

The following list shows minimum operating system and Java versions:

The three certificate authorities are:

Root certificates from the first two authorities are available from Amazon Trust Services, but keeping your computer up-to-date is the more straightforward solution. To learn more about ACM-provided certificates, see AWS Certificate Manager FAQs.