有多种选择，没有一个是简单的“一次性”全文解决方案:

Key name pattern search: Searching for keys starting with some string- if you design key names carefully, then you may have rather quick solution. Search metadata attached to keys: when posting a file to AWS S3, you may process the content, extract some meta information and attach this meta information in form of custom headers into the key. This allows you to fetch key names and headers without need to fetch complete content. The search has to be done sequentialy, there is no "sql like" search option for this. With large files this could save a lot of network traffic and time. Store metadata on SimpleDB: as previous point, but with storing the metadata on SimpleDB. Here you have sql like select statements. In case of large data sets you may hit SimpleDB limits, which can be overcome (partition metadata across multiple SimpleDB domains), but if you go really far, you may need to use another metedata type of database. Sequential full text search of the content - processing all the keys one by one. Very slow, if you have too many keys to process.

几年来，我们每天存储1440个版本的文件(每分钟一个)，使用版本化桶，这是很容易实现的。但要获得一些较旧的版本需要时间，因为人们必须一个版本一个版本地按顺序进行。有时我使用简单的CSV记录索引，显示发布时间和版本id，有了这个，我可以很快跳转到旧版本。

正如你所看到的，AWS S3并不是为全文搜索而设计的，它是一个简单的存储服务。

S3没有原生的“搜索此桶”，因为实际内容是未知的-此外，由于S3是基于键/值的，因此没有原生的方法可以一次访问多个节点，而更传统的数据存储提供了一个(SELECT * FROM…(在SQL模型中)。

您需要做的是执行ListBucket以获得bucket中对象的列表，然后遍历每个项，执行您实现的自定义操作—这就是您的搜索。

虽然不是AWS的原生服务，但有Mixpeek，它在S3文件上运行文本提取，如Tika、Tesseract和ImageAI，然后将它们放在Lucene索引中，使它们可搜索。

在这里查看文档

积分如下:

Download the module: https://github.com/mixpeek/mixpeek-python Import the module and your API keys: from mixpeek import Mixpeek, S3 from config import mixpeek_api_key, aws Instantiate the S3 class (which uses boto3 and requests): s3 = S3( aws_access_key_id=aws['aws_access_key_id'], aws_secret_access_key=aws['aws_secret_access_key'], region_name='us-east-2', mixpeek_api_key=mixpeek_api_key ) Upload one or more existing S3 files: # upload all S3 files in bucket "demo" s3.upload_all(bucket_name="demo") # upload one single file called "prescription.pdf" in bucket "demo" s3.upload_one(s3_file_name="prescription.pdf", bucket_name="demo") Now simply search using the Mixpeek module: # mixpeek api direct mix = Mixpeek( api_key=mixpeek_api_key ) # search result = mix.search(query="Heartgard") print(result) Where result can be: [ { "_id": "REDACTED", "api_key": "REDACTED", "highlights": [ { "path": "document_str", "score": 0.8759502172470093, "texts": [ { "type": "text", "value": "Vetco Prescription\nVetcoClinics.com\n\nCustomer:\n\nAddress: Canine\n\nPhone: Australian Shepherd\n\nDate of Service: 2 Years 8 Months\n\nPrescription\nExpiration Date:\n\nWeight: 41.75\n\nSex: Female\n\n℞ " }, { "type": "hit", "value": "Heartgard" }, { "type": "text", "value": " Plus Green 26-50 lbs (Ivermectin 135 mcg/Pyrantel 114 mg)\n\nInstructions: Give one chewable tablet by mouth once monthly for protection against heartworms, and the treatment and\ncontrol of roundworms, and hookworms. " } ] } ], "metadata": { "date_inserted": "2021-10-07 03:19:23.632000", "filename": "prescription.pdf" }, "score": 0.13313256204128265 } ]

然后解析结果

AWS发布了使用SQL: Amazon Athena https://aws.amazon.com/athena/查询S3桶的新服务

在S3控制台中按前缀搜索

直接在AWS Console桶视图中。

使用s3-dist-cp复制需要的文件

当您有数千或数百万个文件时，另一种获取所需文件的方法是使用分布式复制将它们复制到另一个位置。您可以在Hadoop作业中的EMR上运行此操作。AWS很酷的一点是，他们提供了自定义S3版本S3 -dist-cp。它允许您在groupBy字段中使用正则表达式对需要的文件进行分组。例如，您可以在EMR的自定义步骤中使用它

[
    {
        "ActionOnFailure": "CONTINUE",
        "Args": [
            "s3-dist-cp",
            "--s3Endpoint=s3.amazonaws.com",
            "--src=s3://mybucket/",
            "--dest=s3://mytarget-bucket/",
            "--groupBy=MY_PATTERN",
            "--targetSize=1000"
        ],
        "Jar": "command-runner.jar",
        "Name": "S3DistCp Step Aggregate Results",
        "Type": "CUSTOM_JAR"
    }
]

这不是一个技术性的答案，但我已经构建了一个允许通配符搜索的应用程序:https://bucketsearch.net/

它将异步索引您的bucket，然后允许您搜索结果。

它是免费使用的(捐赠软件)。