add scheme-based caching #94

coreykn · 2020-08-14T04:25:09Z

Cache HTTP and HTTPS pages separately (#20). This builds upon PR #50. This fixes HTTP resources being requested by a page that is loaded over HTTPS. This occurs when a page is loaded over HTTP, cached, and then later loaded over HTTPS. This also fixes HTTPS resources being requested by a page that is loaded over HTTP. If the website does not have a valid SSL/TLS certificate the requests would fail. This issue can arise when HTTP or HTTPS is not being forced.

Update Cache Enabler HTML comment to be more clear on what cached page is being delivered.

Update clear_home() method to unlink based on glob pattern. This builds upon part of PR #64. This is a better way to clear the home page cache as we are not attempting to unlink files that do not exist. Further, this makes clearing the home page cache more simple when introducing scheme-based caching.

Fixes #20

Cache HTTP and HTTPS pages separately (keycdn#20). This builds upon PR keycdn#50. This fixes HTTP resources being requested by a page that is loaded over HTTPS. This occurs when a page is loaded over HTTP, cached, and then later loaded over HTTPS. This also fixes HTTPS resources being requested by a page that is loaded over HTTP. If the website does not have a valid SSL/TLS certificate the requests would fail. This issue can arise when HTTP or HTTPS is not being forced. Update Cache Enabler HTML comment to be more clear on what cached page is being delivered. Update `clear_home()` method to unlink based on glob pattern. This builds upon part of PR keycdn#64. This is a better way to clear the home page cache as we are not attempting to unlink files that do not exist. Further, this makes clearing the home page cache more simple when introducing scheme-based caching. Fixes keycdn#20

Update `Cache_Enabler::process_clear_request()` to only clear the home page cache when using the "Clear URL Cache" admin bar button on the home page. Previously the complete cache would be cleared. Fix advanced cache path to cached variants. This allows the advanced cache to deliver the cached page and bypass PHP as it did in version 1.3.5. This should have been done in PR keycdn#94. Delete `CE_SETTINGS_PATH` constant that was added in PR keycdn#96. The advanced cache is not always able to read the value of this constant.

Update `Cache_Enabler::process_clear_request()` to only clear the home page cache when using the "Clear URL Cache" admin bar button on the home page. Previously the complete cache would be cleared. Update how `Cache_Enabler_Disk::_file_scheme()` returns `https` or `http`. This is required to return the correct scheme if the WordPress installation is behind a proxy. Fix advanced cache path to cached variants. This allows the advanced cache to deliver the cached page and bypass PHP as it did in version 1.3.5. This should have been done in PR keycdn#94. Delete `CE_SETTINGS_PATH` constant that was added in PR keycdn#96. The advanced cache is not always able to read the value of this constant.

Update `Cache_Enabler::process_clear_request()` to only clear the home page cache when using the "Clear URL Cache" admin bar button on the home page. Previously the complete cache would be cleared. Update how `Cache_Enabler_Disk::_file_scheme()` returns `https` or `http`. This is required to return the correct scheme if the WordPress installation is behind a proxy. Fix advanced cache path to cached variants. This allows the advanced cache to deliver the cached page and bypass PHP as it did in version 1.3.5. This should have been done in PR #94. Delete `CE_SETTINGS_PATH` constant that was added in PR #96. The advanced cache is not always able to read the value of this constant.

…123.09beta01 - Cache Enabler plugin v1.4.0 has a caching routine breaking change as they switched to scheme based caching so cached files are prefixed with http- or https- depending on how file is requested keycdn/cache-enabler#94. Nginx served caching looking for cached files in format index.html or index-webp.html prior to v1.4.0. But with v1.4.0 the format cached to http-index.html or https-index.html etc. - Update applies to new centmin.sh menu option 22 cache enabler selected installs and not existing installs. For existing installs, a further update will be made in 123.09beta01 to auto detect existing generated /usr/local/nginx/conf/wpincludes/${vhostname}/wpcacheenabler_${vhostname}.conf include files where ${vhostname} is your wordpress domain name and then auto apply the changes.

Update cache handling to get the cache file based on the request, mostly looking at the Cache Enabler settings and request headers. This updates the cache handling for both new and potentially cached pages by generating the required cache file through cache keys. This will be much better going forward as it simplifies handling current cache variants and adding additional cache variants in the future. The previous way of handling this would mean many methods and convoluted logic would have to be added on each additional cache variant. As each variant is added that would have only lead to more unnecessary complexity. The cache keys are obtained in the following ways: * `scheme`: Scheme-based caching was introduced in PR keycdn#94 and released in version 1.4.0. This was then later updated in PR keycdn#98, PR keycdn#109, and PR keycdn#141. Change this to use what the WordPress `is_ssl()` function uses with the added ability to also check the `X-Forwarded-Proto` or `X-Forwarded-Scheme` request headers. * `device`: Device-based caching will be introduced in this PR and released in version 1.7.0. This will allow a separate mobile cache to be made based on the `User-Agent` request header. The values checked have been taken from the WordPress `wp_is_mobile()` function (keycdn#160). * `webp`: WebP-based caching was released in version 1.0.2. This still checks the `Accept` request header, but will now look for `image/webp` instead of just `webp`. * `compression`: Compression-based caching for Gzip has been supported since the initial release. This still checks the `Accept-Encoding` request header for `gzip`. This setup will make it easy for other compression algorithms to be introduced in the future. This change means that only the requested cache file is created if it does not exist yet instead of all variants at once. This will reduce unnecessary memory usage, especially if many cache variants are used, and will reduce storing unnecessary cache variants that may never even be requested. (This will impact anyone using cache warming techniques because a request for each variant to cache will now have to be made.) This also ensures the client gets what it has requested instead of falling back to the default HTML file if the requested variant does not exist yet but the default HTML file did. (That was pretty rare edge case as it would only previously occur if a cache variant setting was enabled after cached pages were already generated and the cache was not cleared.) Getting the request headers has been updated as in some cases `apache_request_headers()` was a function that existed but returned an empty array or `false`. This lead to some cache variants not being delivered even when accepted by the client. This new way will try to get the request header from `apache_request_headers()` and if not set then fallback to the `$_SERVER` variable. If neither are set then an empty string to prevent having to always check if it is set. Only checking `$_SERVER` variables was not used as some environments do not provide all of these. This way should get the specific request header values we need across many different configurations if it is available. The cache signature will now include the file name, such as `https-index-webp.html.gz`, instead of `https webp gzip` because of this new handling. We can filter this and update it to the old format if preferred. (I do not have a preference, this way just reduces the overhead.) The cache signature will now also include the generated date in HTTP-date format (e.g. `D, d M Y H:i:s GMT`) instead of `d.m.Y H:i:s`. Update `Cache_Enabler_Disk::get_settings_file_name()` to parse the `$_SERVER['REQUEST_URI']` value and get the `PHP_URL_PATH` when in the subdirectory network fallback to prevent a warning from occurring in the following `preg_grep()` function when the path contained a query string. The `?` from the query string would cause a regular expression error if the path had a trailing slash (preceding token is not quantifiable, like in `\.(path|to|page|?param=value)`). For a future note, error handling when creating directories and compressing page contents needs to be improved. `wp_die()` for failed cache file directory creation has been removed as it did not work and would have caused an output buffer error instead. As far as I can tell this would have always occurred in all past versions. This would not occur when trying to create the settings file, but it has been removed for this as well as it is not a graceful way to handle this type of error. If Gzip compression failed the page will not be created. These all rarely occur, if they ever do as no issues have been opened around these scenarios, but in the event they do the user should receive additional insight to what is occurring so it can be resolved. Closes keycdn#160

Update cache handling to get the cache file based on the request, mostly looking at the Cache Enabler settings and request headers. This updates the cache handling for both new and potentially cached pages by generating the required cache file through cache keys. This will be much better going forward as it simplifies handling current cache variants and adding additional cache variants in the future. The previous way of handling this would mean many methods and convoluted logic would have to be added on each additional cache variant. As each variant is added that would have only lead to more unnecessary complexity. The cache keys are obtained in the following ways: * `scheme`: Scheme-based caching was introduced in PR #94 and released in version 1.4.0. This was then later updated in PR #98, PR #109, and PR #141. Change this to use what the WordPress `is_ssl()` function uses with the added ability to also check the `X-Forwarded-Proto` or `X-Forwarded-Scheme` request headers. * `device`: Device-based caching will be introduced in this PR and released in version 1.7.0. This will allow a separate mobile cache to be made based on the `User-Agent` request header. The values checked have been taken from the WordPress `wp_is_mobile()` function (#160). * `webp`: WebP-based caching was released in version 1.0.2. This still checks the `Accept` request header, but will now look for `image/webp` instead of just `webp`. * `compression`: Compression-based caching for Gzip has been supported since the initial release. This still checks the `Accept-Encoding` request header for `gzip`. This setup will make it easy for other compression algorithms to be introduced in the future. This change means that only the requested cache file is created if it does not exist yet instead of all variants at once. This will reduce unnecessary memory usage, especially if many cache variants are used, and will reduce storing unnecessary cache variants that may never even be requested. (This will impact anyone using cache warming techniques because a request for each variant to cache will now have to be made.) This also ensures the client gets what it has requested instead of falling back to the default HTML file if the requested variant does not exist yet but the default HTML file did. (That was pretty rare edge case as it would only previously occur if a cache variant setting was enabled after cached pages were already generated and the cache was not cleared.) Getting the request headers has been updated as in some cases `apache_request_headers()` was a function that existed but returned an empty array or `false`. This lead to some cache variants not being delivered even when accepted by the client. This new way will try to get the request header from `apache_request_headers()` and if not set then fallback to the `$_SERVER` variable. If neither are set then an empty string to prevent having to always check if it is set. Only checking `$_SERVER` variables was not used as some environments do not provide all of these. This way should get the specific request header values we need across many different configurations if it is available. The cache signature will now include the file name, such as `https-index-webp.html.gz`, instead of `https webp gzip` because of this new handling. We can filter this and update it to the old format if preferred. (I do not have a preference, this way just reduces the overhead.) The cache signature will now also include the generated date in HTTP-date format (e.g. `D, d M Y H:i:s GMT`) instead of `d.m.Y H:i:s`. Update `Cache_Enabler_Disk::get_settings_file_name()` to parse the `$_SERVER['REQUEST_URI']` value and get the `PHP_URL_PATH` when in the subdirectory network fallback to prevent a warning from occurring in the following `preg_grep()` function when the path contained a query string. The `?` from the query string would cause a regular expression error if the path had a trailing slash (preceding token is not quantifiable, like in `\.(path|to|page|?param=value)`). For a future note, error handling when creating directories and compressing page contents needs to be improved. `wp_die()` for failed cache file directory creation has been removed as it did not work and would have caused an output buffer error instead. As far as I can tell this would have always occurred in all past versions. This would not occur when trying to create the settings file, but it has been removed for this as well as it is not a graceful way to handle this type of error. If Gzip compression failed the page will not be created. These all rarely occur, if they ever do as no issues have been opened around these scenarios, but in the event they do the user should receive additional insight to what is occurring so it can be resolved. Closes #160

coreykn requested a review from svenba August 14, 2020 04:25

svenba approved these changes Aug 14, 2020

View reviewed changes

svenba merged commit be62c3d into keycdn:master Aug 14, 2020

coreykn deleted the add-scheme-based-caching branch August 14, 2020 15:15

coreykn mentioned this pull request Aug 19, 2020

update clear requests and fix advcache #98

Merged

centminmod mentioned this pull request Aug 27, 2020

scheme-based cache file generation bug in v1.4.6 #108

Closed

coreykn mentioned this pull request Mar 12, 2021

update cache handling #211

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

add scheme-based caching #94

add scheme-based caching #94

coreykn commented Aug 14, 2020

add scheme-based caching #94

add scheme-based caching #94

Conversation

coreykn commented Aug 14, 2020