Google Search Appliance software version 6.0
Posted June 2009
Crawling is the process where the Google Search Appliance discovers enterprise content to index. This chapter provides an overview of how the Google Search Appliance crawls public content.
Before anyone can use the Google Search Appliance to search your enterprise content, the search appliance must build the search index, which enables search queries to be quickly matched to results. To build the search index, the search appliance must browse, or "crawl" your enterprise content, as illustrated in the following example.
The administration at Missitucky University plans to offer its staff, faculty, and students simple, fast, and secure search across all their content using the Google Search Appliance. To achieve this goal, the search appliance must crawl their content, starting at the Missitucky University Web site's home page.
Missitucky University has a Web site that provides categories of information such as Admissions, Class Schedules, Events, and News Stories. The Web site's home page lists hyperlinks to other URLs for pages in each of these categories. For example, the News Stories hyperlink on the home page points to a URL for a page that contains hyperlinks to all recent news stories. Similarly, each news story contains hyperlinks that point to other URLs.
The relations among the hyperlinks within the Missitucky University Web site constitute a virtual web, or pathway that connects the URLs to each other. Starting at the home page and following this pathway, the search appliance can crawl from URL to URL, browsing content as it goes.
Crawling Missitucky University's content actually begins with a list of URLs ("start URLs") where the search appliance should start browsing; in this example, the first start URL is the Missitucky University home page.
The search appliance visits the Missitucky University home page, then it:
By repeating these steps for each URL in the crawl queue, the search appliance can crawl all of Missitucky University's content. As a result, the search appliance gathers the information that it needs to build the search index, and ultimately, to serve search results to end users.
Because Missitucky University's content changes constantly, the search appliance continuously crawls it to keep the search index and the search results up-to-date.
The Google Search Appliance supports two modes of crawling:
For information about choosing a crawl mode and starting a crawl see Selecting a Crawl Mode.
In continuous crawl mode, the search appliance is crawling your enterprise content at all times, ensuring that newly added or updated content is added to the index as quickly as possible. After the Google Search Appliance is installed, it defaults to continuous crawl mode and establishes the default collection.
The search appliance can automatically determine URLs that often change and should be crawled frequently and URLs that seldom change and should be crawled infrequently. Each URL is crawled twice its "change interval," which is how often it is observed to change. For example, if your corporate intranet portal page changes every 24 hours, the search appliance crawls it every 12 hours.
The search appliance does not recrawl any URLs until all new URLs have been discovered or the license limit has been reached. A URL in the index is recrawled even if there are no longer any links to that URL from other pages in the index.
In scheduled crawl mode, the Google Search Appliance crawls your enterprise content at a scheduled time.
The Google Search Appliance can crawl and index content that is stored in the following types of sources:
Crawling FTP is not supported on the Google Search Appliance.
Public Web content is available to all users. The Google Search Appliance can crawl and index both public and secure enterprise content that resides on a variety of Web servers, including these:
Secure Web content is protected by authentication mechanisms and is available only to users who are members of certain authorized groups. The Google Search Appliance can crawl and index secure content protected by:
The search appliance can crawl and index content protected by forms-based single sign-on systems.
For HTTPS websites, the Google Search Appliance uses a serving certificate as a client certificate when crawling. You can upload a new serving certificate using the Admin Console. Some Web servers do not accept client certificates unless they are signed by trusted Certificate Authorities.
The Google Search Appliance can also crawl several file formats, including Microsoft Word, Excel, and Adobe PDF that reside on network file shares. The crawler can access content over Server Message Block (SMB) protocol (the standard network file share protocol on Microsoft Windows, supported by the SAMBA server software and numerous storage devices).
For a complete list of supported file formats, refer to Indexable File Formats.
The Google Search Appliance can crawl databases directly. To access content in a database, the Google Search Appliance sends SQL (Structured Query Language) queries using JDBC (Java Database Connectivity) adapters provided by each database company.
For information about crawling databases, refer to Database Crawling and Serving.
The Google Search Appliance does not crawl or index enterprise content that is excluded by these mechanisms:
Also the Google Search Appliance cannot:
The following sections describe all these exclusions.
A Google Search Appliance administrator can prohibit the crawler from following and indexing particular URLs. For example, any URL that should not appear in search results or be counted as part of the search appliance license limit should be excluded from crawling. For more information, refer to Configuring a Crawl.
To prohibit any crawler from accessing all or some of the content on an HTTP or HTTPS site, a content server administrator or webmaster typically adds a robots.txt file to the content server or Web site. This file tells the crawlers to ignore all or some files and directories on the server or site. Documents crawled using other protocols, such as SMB, are not affected by the restrictions of robots.txt. For the Google Search Appliance to be able to access the robot.txt file, the file must be public.
The Google Search Appliance crawler always obeys the rules in robots.txt. You cannot override this feature. Before crawling HTTP or HTTPS URLs on a host, a Google Search Appliance fetches the robots.txt file. For example, before crawling any URLs on http://www.mycompany.com/ or https://www.mycompany.com/, the search appliance fetches http://www.mycompany.com/robots.txt.
When the search appliance requests the robots.txt file, the host returns an HTTP response that determines whether or not the search appliance can crawl the site. The following table lists HTTP responses and how the Google Search Appliance crawler responds to them.
| HTTP Response | File Returned? | Google Search Appliance Crawler Response |
|---|---|---|
| 200 OK |
Yes | The search appliance crawler obeys exclusions specified by robots.txt when fetching URLs on the site. |
| 404 Not Found |
No | The search appliance crawler assumes that there are no exclusions to crawling the site and proceeds to fetch URLs. |
| Other responses | The search appliance crawler assumes that it is not permitted to crawl the site and does not fetch URLs. |
When crawling, the search appliance caches robots.txt files and refetches a robots.txt file if 30 minutes have passed since the previous fetch. If changes to a robots.txt file prohibit access to documents that have already been indexed, those documents are removed from the index. If the search appliance can no longer access robots.txt on a particular site, all the URLs on that site are removed from the index.
For general information on how robots.txt files, as well as how Robots META tags work, see the Robots exclusion standard at: http://www.robotstxt.org/wc/exclusion.html. For detailed information about HTTP status codes, click here.
The Google Search Appliance does not crawl a Web page if it has been marked with the nofollow Robots META tag.
The Google Search Appliance does not crawl links that are embedded within an area tag. The HTML area tag is used to define a mouse-sensitive region on a page, which can contain a hyperlink. When the user moves the pointer into a region defined by an area tag, the arrow pointer changes to a hand and the URL of the associated hyperlink appears at the bottom of the window.
For example, the following HTML defines an region that contains a link:
<map
name="n5BDE56.Body.1.4A70"> <area shape="rect" coords="0,116,311,138" id="TechInfoCenter" href="http://www.bbb.com/main/help/ourcampaign/ourcampaign.htm" alt=""></map>
When the search appliance crawler follows newly discovered links in URLs, it does not follow the link (http://www.bbb.com/main/help/ourcampaign/ourcampaign.htm) within this area tag.
Because the Google Search Appliance crawler discovers new content by following links within documents, it cannot find a URL that is not linked from another document through this process.
You can enable the search appliance crawler to discover any unlinked URLs in your enterprise content by:
Before crawling starts, the Google Search Appliance administrator configures the crawl path, which includes URLs where crawling should start, as well as URL patterns that the crawler should follow and should not follow. Other information that webmasters, content owners, and search appliance administrators typically prepare before crawling starts includes:
This section describes how the Google Search Appliance crawls Web and network file share content as it applies to both scheduled crawl and continuous crawl modes.
This section contains data flow diagrams, used to illustrate how the Google Search Appliance crawls enterprise content. The following table describes the symbols used in these diagrams.
| Symbol | Definition | Example |
|---|---|---|
 |
Start state or Stop state | Start crawl, end crawl |
 |
Process | Follow links within the document |
 |
Data store, which can be a database, file system, or any other type of data store | Crawl queue |
 |
Data flow among processes, data stores, and external interactors | URLs |
 |
External input or terminator, which can be a process in another diagram | Delete URL |
 |
Callout to a diagram element |
The following diagram provides an overview of the following major crawling processes:
The sections following the diagram provide details about each of the these major processes.
The crawl queue is a list of URLs that the Google Search Appliance will crawl. The search appliance associates each URL in the crawl queue with a priority, typically based on estimated Enterprise PageRank. Enterprise PageRank is a measure of the relative importance of a Web page within the set of your enterprise content. It is calculated using a link-analysis algorithm similar to the one used to calculate PageRank on google.com.
The order in which the Google Search Appliance crawls URLs is determined by the crawl queue. The following table gives an overview of the priorities assigned to URLs in the crawl queue.
| Source of URL | Basis for Priority |
|---|---|
| Start URLs (highest) | Fixed priority |
| New URLs that have never been crawled | Estimated Enterprise PageRank |
| Newly discovered URLs | For a new crawl, estimated Enterprise PageRank |
| For a recrawl, estimated Enterprise PageRank and a factor that ensures that new documents are crawled before previously indexed content | |
| URLs that are already in the index (lowest) | Enterprise PageRank, the last time it was crawled, and estimated change frequency |
By crawling URLs in this priority, the search appliance ensures that the freshest, most relevant enterprise content appears in the index.
Tip: Although it is not possible to view the Enterprise PageRank for a URL, you can view the RK (Ranking Value) tag in XML search results. The RK tag is an approximate indicator of the Enterprise PageRank of a URL. To request XML search results, use the output parameter in a search request, for example: output=xml.
After configuring the crawl path and preparing content for crawling, the search appliance administrator starts a continuous or scheduled crawl. The following diagram provides an overview of starting the crawl and populating the crawl queue.
When crawling begins, the search appliance populates the crawl queue with URLs. The following table lists the contents of the crawl queue for a new crawl and a recrawl.
| Type of Crawl | Crawl Queue Contents |
|---|---|
| New crawl | The start URLs that the search appliance administrator has configured. |
| Recrawl | The start URLs that the search appliance administrator has configured and the complete set of URLs contained in the current index. |
The Google Search Appliance crawler attempts to fetch the URL with the highest priority in the crawl queue. The following diagram provides an overview of this process.
If the search appliance successfully fetches a URL, it downloads the document and uses the checksum to test if there are already 100 documents with the same checksum in the index (this limit is subject to change in future search appliance releases). If there are 100 documents with the same checksum in the index, the document is considered a duplicate and discarded (in Crawl Diagnostics, the document is shown as "Considered Duplicate"). If there are not 100 documents with the same checksum in the index, the search appliance caches the document for indexing.
Generally, if the search appliance fails to fetch a URL, it deletes the URL from the crawl queue. Depending on several factors, the search appliance may take further action when it fails to fetch a URL.
When fetching documents from a slow server, the search appliance paces the process so that it does not cause server problems. The search appliance administrator can also adjust the number of concurrent connections to a server by configuring the web server host load schedule.
During the recrawl of an indexed document, the Google Search Appliance sends the If-Modified-Since header based on the last crawl date of the document. If the web server returns a 304 Not Modified response, the appliance does not further process the document. If the web server returns content, the Google Search Appliance computes the checksum of the newly downloaded content and compares it to the checksum of the previous content. If the checksum is the same, then appliance does not further process the document.
To detect changes to cached documents when recrawling it, the search appliance:
When the Google Search Appliance fetches a URL from a file share, the object that it actually retrieves and the method of processing it depends on the type of object that is requested. For each type of object requested, the following table provides an overview of the process that the search appliance follows. For information on how these objects are counted as part of the search appliance license limit, refer to When Is a Document Counted as Part of the License Limit?
| Requested Object | Google Search Appliance Process Overview |
|---|---|
| Document |
|
| Directory |
|
| Share |
|
| Host |
|
When the Google Search Appliance fetches a document, it determines the size and type of the the file. If the file is:
For each document that it indexes, the Google Search Appliance follows newly discovered URLs (HTML links) within that document.
Before following a newly discovered link, the search appliance checks the URL against:
If the URL passes these checks, the search appliance adds the URL to the crawl queue, and eventually crawls it. If the URL does not pass these checks, the search appliance deletes it from the crawl queue. The following diagram provides an overview of this process.
The search appliance crawler only follows HTML links in the following format:
<a href="/page2.html">link to page 2</a>
It follows HTML links in PDF files, Word documents, and Shockwave documents. The search appliance crawler does not follow HTML links embedded in Javascript code.
The Google Search Appliance administrator can end a continuous crawl by pausing it.
The search appliance administrator can configure a scheduled crawl to end at a specified time. A scheduled crawl also ends when the license limit is reached. The following table provides more details about the conditions that cause a scheduled crawl to end.
| Condition | Description |
|---|---|
| Scheduled end time | Crawling stops at its scheduled end time. |
| Crawl to completion | There are no more URLs in the crawl queue. The search appliance crawler has discovered and attempted to fetch all reachable content that matches the configured URL patterns. |
| The license limit is reached | The search appliance license limits the maximum number of URLs in the index. When the search appliance reaches this limit, it stops crawling new URLs. The search appliance removes the excess URLs from the crawl queue. |
For both scheduled crawls and continuous crawls, documents usually appear in search results approximately 30 minutes after they are crawled. This period can increase if the system is under a heavy load, or if there are many non-HTML documents.
For a recrawl, if an older version of a document is cached in the index from a previous crawl, the search results refer to the cached document until the new version is available.
The search appliance determines the priority of URLs for recrawl using the following rules, listed in order from highest to lowest priority:
There are some other factors that also contribute to whether a URL is recrawled, for example how fast a host can respond will also play a factor, or whether it received an error on the last crawl attempt.
To see how often a URL has been recrawled in the past, as well as the status of the URL, you can view the crawl history of a single URL by using the Status and Reports > Crawl Diagnostics page in the Admin Console.
When crawling, the Google Search Appliance tests network connectivity by attempting to fetch every start URL every 30 minutes. If approximately 10% of the start URLs return HTTP 200 (OK) responses, the search appliance assumes that there are no network connectivity issues. If less than 10% return OK responses, the search appliance assumes that there are network connectivity issues with a content server and slows down or stops.
During a temporary network outage, slowing or stopping a crawl prevents the search appliance from removing URLs that it cannot reach from the index. The crawl speeds up or restarts when the start URL connectivity test returns an HTTP 200 response.
Your Google Search Appliance license determines the number of documents that can appear in your index, as listed in the following table.
| Search Appliance Model | Maximum License Limit |
|---|---|
| GB-1001 | 3 million |
| GB-7007 | 10 million |
| GB-9009 | 30 million |
For a Google Search Appliance, between 3 million and 30 million documents can appear in the index, depending on your model.
For example, if the license limit is 3 million, the search appliance crawler attempts to put the 3 million documents in the index. During a recrawl, when the crawler discovers a new URL, it must decide whether to crawl the document.
When the search appliance reaches its limit, it stops crawling new URLs, and removes documents from the index to bring the total number of documents to the license limit.
Google recommends managing crawl patterns on the Crawl and Index > Crawl URLs page in the Admin Console to ensure that the total number of URLs that match the crawl patterns remains at or below the license limit.
Generally, when the Google Search Appliance successfully fetches a document, it is counted as part of the license limit. If the search appliance does not successfully fetch a document, it is not counted as part of the license limit. The following table provides an overview of the conditions that determine whether or not a document is counted as part of the license limit.
| Condition | Counted as Part of the License Limit? |
|---|---|
| The search appliance fetches a URL without errors. This includes HTTP responses 200 (success), 301 (redirect, URL moved permanently), 302 (redirect, URL moved temporarily), and 304 (not modified) |
The URL is counted as part of the license limit. |
| The search appliance cannot fetch a URL. Instead, the search appliance receives an HTTP error response, such as 404 (document not found) or 500 (temporary server error). | The URL is not counted as part of the license limit. |
| The search appliance fetches two URLs that contain exactly the same content without errors. | Both URLs are counted as part of the license limit, but the one with the lower Enterprise PageRank is automatically filtered out of search results. It is not possible to override this automatic filtering. |
| The search appliance fetches a document from a file share. | The document is counted as part of the license limit. |
| The search appliance retrieves a list of files and subdirectories and in a file share and converts it to a directory listings page. | Each directory in the list is counted as part of the license limit, even if the directory is empty. |
| The search appliance retrieves a list of file shares on a host and converts it to a share listings page. | Each share in the list is counted as part of the license limit. |
To view license information for your Google Search Appliance, use the Administration>License page in the Admin Console.
The Google Search Appliance crawler stores a maximum number of URLs that can be crawled. The maximum number depends on the search appliance model and license limit, as listed in the following table.
| Search Appliance Model | Maximum License Limit | Maximum Number of URLs that Match Crawl Patterns |
|---|---|---|
| GB-1001 | 3 million | ~ 5.2 million |
| GB-7007 | 10 million | ~ 13.6 million |
| GB-9009 | 30 million | ~ 40 million |
If the Google Search Appliance has reached the maximum number of URLs that can be crawled, this number appears in URLs Found That Match Crawl Patterns on the Status and Reports > Crawl Status page in the Admin Console.
Once the maximum number is reached, a new URL is considered for crawling only if it has a higher priority than the least important known URL. In this instance, the higher priority URL is crawled and the lower priority URL is discarded.
For an overview of the priorities assigned to URLs in the crawl queue, see Starting the Crawl and Populating the Crawl Queue.
To enable search results to be sorted and presented based on dates, the Google Search Appliance extracts dates from documents according to rules configured by the search appliance administrator.
In Google Search Appliance software version 4.4.68 and later, document dates are extracted from Web pages when the document is indexed.
The search appliance extracts the first date for a document with a matching URL pattern that fits the date format associated with the rule. If a date is written in an ambiguous format, the search appliance assumes that it matches the most common format among URLs that match each rule for each domain that is crawled. For this purpose, a domain is one level above the top level. For example, mycompany.com is a domain, but intranet.mycompany.com is not a domain.
The search appliance periodically runs a process that calculates which of the supported date formats is the most common for a rule and a domain. After calculating the statistics for each rule and domain, the process may modify the dates in the index. The process first runs 12 hours after the search appliance is installed, and thereafter, every seven days. The process also runs each time you change the document date rules.
The search appliance will not change which date is most common for a rule until after the process has run. Regardless of how often the process runs, the search appliance will not change the date format more than once a day. The search appliance will not change the date format unless 5,000 documents have been crawled since the process last ran.
If you import a configuration file with new document dates after the process has first run, then you may have to wait at least seven days for the dates to be extracted correctly. The reason is that the date formats associated with the new rules are not calculated until the process runs. If no dates were found the first time the process ran, then no dates are extracted until the process runs again.
If no date is found, the search appliance indexes the document without a date.
Normally, document dates appear in search results about 30 minutes after they are extracted. In larger indexes, the process can several hours to complete because the process may have to look at the contents of every document.
The Google Search Appliance index includes all the documents it has crawled. These documents remain in the index and the search appliance continues to crawl them until either one of the following conditions is true:
The search appliance administrator can also remove documents from the index manually.
Removing all links to a document in the index does not remove the document from the index.
The following table describes the conditions that cause documents to be removed from the index.
| Condition | Description |
|---|---|
| The license limit is exceeded | The limit on the number of URLs in the index is the value of Maximum number of pages overall on the Administration > License page. |
| The crawl pattern is changed | To determine which content should be included in the index, the search appliance uses the follow and crawl and do not crawl URL patterns specified on the Crawl and Index > Crawl URLs page. If these URL patterns are modified, the search appliance examines each document in the index to determine whether it should be retained or removed. If the URL does not match any follow and crawl patterns, or if it matches any do not crawl patterns, it is removed from the index. Document URLs disappear from search results between 15 minutes and six hours after the pattern changes, depending on system load. |
| The robots.txt file is changed | If the robots.txt file for a content server or web site has changed to prohibit search appliance crawler access, URLs for the server or site are removed from the index. |
| Authentication failure (401) | If the search appliance receives three successive 401 (authentication failure) errors from the Web server when attempting to fetch a document, the document is removed from the index after the third failed attempt. |
| Document is not found (404) | If the search appliance receives a 404 (Document not found) error from the Web server when attempting to fetch a document, the document is removed from the index. |
Note: Search appliance software versions prior to 4.6 include a process called the "remove doc ripper." This process removes documents from the index every six hours. If the appliance has crawled more documents than its license limit, the ripper removes documents that are below the Enterprise PageRank threshold. The ripper also removes documents that don't match any follow patterns or that do match exclude patterns. If you want to remove documents from search results, use the Remove URLs feature on the Serving > Front Ends > Remove URLs page. When the remove doc ripper has run with your changes to the crawl patterns, you should delete all Remove URL patterns. The Remove URL patterns are checked at search query time and are expensive to process. A large number of Remove URLs patterns affects search query speed.
|
