Slice Plugin

This plugin takes client requests and breaks them up into successive aligned block requests. This supports both whole asset and single range requests.

Purpose

This slice plugin, along with the cache_range_requests plugin allows the following:

  • Fulfill arbitrary range requests by fetching a minimum number of cacheable aligned blocks to fulfill the request.

  • Breaks up very large assets into much smaller cache blocks that can be spread across multiple storage devices and within cache groups.

How It Works

The slice plugin takes GET requests and breaks them into successive aligned range requested blocks. It issues these range requests back into the ATS instance and relies on the cache_range_requests plugin to interact with the caching layer.

This design was chosen because the cache_range_requests plugin was already proven to work well with the ATS caching layer. The slice plugin has the already difficult task of managing multiple block requests and all of the associated flow control between the cache_range_requests plugin and the downstream clients.

Configuration

This plugin is intended for use as a remap plugin and is configured in remap.config.

Or preferably per remap rule in remap.config:

map http://ats/ http://parent/ @plugin=slice.so \
    @plugin=cache_range_requests.so

In this case, the plugin will use the default behaviour:

  • Fulfill whole file or range requests by requesting cacheable block aligned ranges from the parent and assemble them into client responses, either 200 or 206 depending on the client request.

  • Default block size is 1mb (1048576 bytes).

  • This plugin depends on the cache_range_requests plugin to perform actual parent fetching and block caching and If-* conditional header evaluations.

Plugin Options

The slice plugin supports the following options:

--blockbytes=<bytes> (optional)
    Default is 1m or 1048576 bytes
    -b <bytes> for short.
    Suffix k,m,g supported
    Limited to 32k and 128m inclusive.

--blockbytes-test=<bytes> (optional)
    Suffix k,m,g supported
    -t <bytes> for short.
    Limited to any positive number.
    Ignored if --blockbytes provided.

--disable-errorlog (optional)
    Disable writing block stitch errors to the error log.
    -d for short

--exclude-regex=<regex> (optional)
    If provided, only slice what matches.
    If not provided will always slice
    Cannot be used with --include-regex
    -e for short

--include-regex=<regex> (optional)
    If provided, only slice what matches.
    If not provided will always slice
    Cannot be used with --exclude-regex
    -i for short

--pace-errorlog=<seconds> (optional)
    Limit stitching error logs to every 'n' second(s)
    -p for short

--ref-relative (optional)
    Self healing mode typically uses slice 0 as the reference slice
    for every request.  This is very safe but also increases plugin
    time and latency as the first slice is always fully processed
    whether or not the original requests needs any data from slice 0.
    This option uses the first slice in the request as reference
    which has better performance.  A downside of this mode is that
    self healing won't happen if blocks in a request agree.
    Normally leave this off.
    -l for short

--remap-host=<loopback hostname> (optional)
    Uses effective url with given hostname for remapping.
    Requires setting up an intermediate loopback remap rule.
    -r for short

--skip-header=<header name> (default: X-Slicer-Info)
    Header name used by the slice plugin after the loopback
    to indicate that the slice plugin should be skipped.
    -s for short

--crr-ident-header=<header name> (default: X-Crr-Ident)
    Header name used by the slice plugin to tell the
    `cache_range_requests` plugin the identifier of the
    first/reference slice fetched.  First Etag is preferred
    followed by Last-Modified. The `cache_range_requests`
                            plugin uses this header to flip cache lookup status
                            to STALE or FRESH depending on the header.

--prefetch-count=<int> (optional)
    Default is 0
    Prefetches successive 'n' slice block requests in the background
    and caches (with `cache_range_requests` plugin). Prefetching is only
    enabled when first block (of the client request) is a cacheable object
    with miss or hit-stale status. Especially for large objects, prefetching
    can improve cache miss latency.
    -f for short

--strip-range-for-head (optional)
    Enable slice plugin to strip Range header for HEAD requests.
    -h for short

--minimum-size (optional)
--metadata-cache-size (optional)
--stats-prefix (optional)
    In combination, these three options allow for conditional slice.
    Specify the minimum size object to slice with --minimum-size.  Allowed
    values are the same as --blockbytes.  Conditional slicing uses a cache
    of object sizes to make the decision of whether to slice.  The cache
    will only store the URL of large objects as they are discovered in
    origin responses.  You should set the --metadata-cache-size to by
    estimating the working set size of large objects.  You can use
    stats to determine whether --metadata-cache-size was set optimally.
    Stat names are prefixed with the value of --stats-prefix.  The names
    are:

    <prefix>.metadata_cache.true_large_objects - large object cache hits
    <prefix>.metadata_cache.true_small_objects - small object cache hits
    <prefix>.metadata_cache.false_large_objects - large object cache misses
    <prefix>.metadata_cache.false_small_objects - small object cache misses
    <prefix>.metadata_cache.no_content_length - number of responses without content length
    <prefix>.metadata_cache.bad_content_length - number of responses with invalid content length
    <prefix>.metadata_cache.no_url - number of responses where URL parsing failed

    If an object size is not found in the object size cache, the plugin
    will not slice the object, and will turn off ATS cache on this request.
    The object size will be cached in following requests, and slice will
    proceed normally if the object meets the minimum size requirement.

    Range requests from the client for small objects are passed through the
    plugin unchanged.  If you use the `cache_range_requests` plugin, slice plugin
    will communicate with `cache_range_requests` using an internal header
    that causes `cache_range_requests` to be bypassed in such requests, and
    allow ATS to handle those range requests internally.

--purge-probe-blocks=<int> (optional)
    Default is 8
    How many consecutive uncached slice blocks a PURGE walks, before any block
    has reported the object's extent, until it concludes that nothing about the
    object is cached.  May be overridden per request with the header named by
    ``--purge-probe-header``.  See `Purge Requests`_.
    -q for short

--purge-probe-header=<string> (optional)
    Default is X-Slice-Purge-Probe
    Name of the request header a PURGE may use to override
    ``--purge-probe-blocks`` for that request.  A malformed value is ignored
    in favour of the configured default.  Slice strips this header from the
    block requests it issues.
    -H for short

Examples:

@plugin=slice.so @pparam=--blockbytes=1000000 @plugin=cache_range_requests.so

Or alternatively:

@plugin=slice.so @pparam=-b @pparam=1000000 @plugin=cache_range_requests.so

Byte suffix examples:

slice.so --blockbytes=5m
slice.so -b 512k
slice.so --blockbytes=32m

For testing and extreme purposes the parameter blockbytes-test may be used instead which is unchecked:

slice.so --blockbytes-test=1G
slice.so -t 13

Because the slice plugin is susceptible to errors during block stitching extra logs related to stitching are written to diags.log. Worst case an error log entry could be generated for every transaction. The following options are provided to help with log overrun:

slice.so --pace-errorlog=5
slice.so -p 1
slice.so --disable-errorlog

After modifying remap.config, restart or reload Traffic Server (sudo traffic_ctl config reload) or (sudo traffic_ctl server restart) to activate the new configuration values.

Don’t slice txt files:

slice.so --exclude-regex=\\.txt
slice.so -e \\.txt

Slice only mp4 files:

slice.so --include-regex=\\.mp4
slice.so -i \\.mp4

Debug Options

While the current slice plugin is able to detect block consistency errors during the block stitching process, it can only abort the client connection. A CDN can only “fix” these by issuing an appropriate content revalidation.

Under normal logging these slice block errors tend to show up as:

pscl value 0
crc value ERR_READ_ERROR

By default more detailed stitching errors are written to diags.log.

Whether or how often these detailed log entries are written are configurable plugin options.

Implementation Notes

This slice plugin is a stop gap plugin for handling special cases involving very large assets that may be range requested. Hopefully the slice plugin is deprecated in the future when partial object caching is finally implemented.

Slice ONLY handles slicing up requests into blocks, it delegates actual caching and fetching to the cache_range_requests.so plugin.

Plugin Function

Below is a quick functional outline of how a request is served by a remap rule containing the Slice plugin with cache_range_requests:

For each client request that comes in all remap plugins are run up until the slice plugin is hit. If the slice plugin can be run (ie: GET request) it will handle the request and STOP any further plugins from executing.

At this point the request is sliced into 1 or more blocks by adding in range request headers (“Range: bytes=”). A special header X-Slicer-Info header is added and the pristine URL is restored.

For each of these blocks separate sequential TSHttpConnect(s) are made back into the front end of ATS. By default of the remap plugins are rerun. Slice skips the remap due to presence of the X-Slicer-Info header and allows cache_range_requests.so to serve the slice block back to Slice either via cache OR parent request.

Slice assembles a header based on the very first slice block response and sends it to the client. If necessary it then skips over bytes in the first block and starts sending byte content, examining each block header and sends its bytes to the client until the client request is satisfied.

Any extra bytes at the end of the last block are consumed by the Slice plugin to allow cache_range_requests to finish the block fetch to ensure the block is cached.

Self Healing

The slice plugin uses the very first slice as a reference slice which uses content-length and etag or last-modified headers to ensure assembled blocks come from the same asset. In the case where a slice from a parent is fetched which indicates that the asset has changed, the slice plugin will attempt to self heal the asset. The cache_range_requests plugin must be configured with the –consider-ident parameter in order for this to work.

Example remap.config configuration:

map http://slice/ http://parent/ @plugin=slice.so @pparam=--remap-host=cache_range_requests
map http://cache_range_requests/ http://parent/ @plugin=cache_range_requests.so @pparam=--consider-ident

When a request is served, the slice plugin uses the header from slice 0 requested range build a response to the client. When subsequent slices are requested from the parent the X-Crr-Ident header is populated with the reference identifier (etag or last-modified) and the request is made through the cache_range_requests plugin. The cache_range_requests plugin will then decide whether to send back the current cached slice (if the identifier matches) or attempt to refetch or “heal” that slice by marking it STALE.

If the slice returned by the cache_range_requests plugin still doesn’t match the reference slice then client side of the transaction is aborted. The plugin will then attempt to “heal” the reference slice. The X-Crr-Ident header is populated with the new identifer and the reference slice is re-requested with the intent of having the cache_range_requests plugin “heal” the reference slice.

The plugin may be configured to use the first slice of the request as the reference slice instead of the asset slice 0. This option is faster as it does not visit any slices outside those needed to fulfill a request. This option may cause serious out of sync issues as range requests may end up being served from temporally different assets.

Purge Requests

The slice plugin supports PURGE requests, discarding the requested object from cache. Without a range every block is discarded; with a range, the blocks that range covers are. Two cases below purge more than the range names: a suffix range, and block 0 when --ref-relative is disabled.

Slice issues one PURGE per block and walks every block it was asked for, whether or not each one is currently cached. A block that is already absent answers 404 internally; that is simply noted and the walk continues, so a gap left by per-block eviction cannot leave the blocks behind it in cache.

Slice learns where the object ends from the blocks it removes. PURGE is a Traffic Server extension, so a successful block purge reports the removed object’s extent in a X-Purged-Content-Range header, and the walk continues to the last block that extent implies. Blocks of one object can disagree about its length when the origin object has been replaced in place; slice takes the largest extent any block reports, so the longer generation’s tail is not left behind.

Until some block has reported an extent, a walk over an open-ended range has no end but the miss bound: it stops after --purge-probe-blocks consecutive uncached blocks and reports that nothing was found. That is what bounds a PURGE for a URL which is not cached at all.

An operator often knows more about the object than the plugin does, since the block count is just the object’s size divided by the block size. That count can be supplied per request with the header named by --purge-probe-header, default X-Slice-Purge-Probe:

PURGE /obj HTTP/1.1
X-Slice-Purge-Probe: 64

This only changes how long the walk keeps going without having found anything; it never limits how many blocks are purged once an extent is known.

The bound has to be able to span a whole object, because in the worst case only the object’s last block is still cached, so the value an operator wants is the object’s size divided by the block size: a 10 GB object in 1 MB blocks needs 10240. There is no ceiling on it beyond that, since reaching the bound costs one internal cache lookup per block and PURGE is already restricted by ip_allow.yaml. A malformed value is ignored in favour of the configured default, and slice strips the header from the block requests it issues.

If the bound is reached, slice logs that it gave up and reports 404 even though later blocks may still be cached. Raise --purge-probe-blocks, or send the override, for objects whose leading blocks are routinely absent.

A client range that is already closed, such as bytes=0-6399999999, bounds the walk directly, and is clamped against the object’s extent as soon as some block reports one. An over-estimate therefore costs no extra block PURGEs beyond the end of the object, and if no block is cached at all the miss bound stops the walk.

A suffix range, bytes=-<n>, names its blocks by their distance from an end slice does not know yet, and purging is the only way it could find out. Rather than guess at the start, such a purge is widened to the whole object: a superset of what was asked for, so the named blocks certainly go. A GET with the same header is unaffected and still returns exactly the last n bytes.

The response is sent once the walk is complete: 200 if at least one block was removed, 404 if none was found. This matches what Traffic Server reports for a PURGE of an object that is not sliced.

A block whose PURGE returns neither 200 nor 404 — a 403 from ip_allow.yaml, for instance, or a 502 — says nothing about whether that block was cached, and nothing about the blocks behind it. The walk stops there and that status is reported in place of 200, so the two success statuses keep meaning what they say: 200 that the object is gone and 404 that it was not there, never that this proxy could not tell. Blocks the walk had already removed stay removed, and blocks behind the failing one are left cached, so such a PURGE is worth repeating once whatever refused it has been dealt with.

The functionality works with --ref-relative both enabled and disabled. With it disabled, block 0 is always the first block walked, so a PURGE whose range does not cover block 0 still purges it.

Conditional Slicing

The goal of conditional slicing is to slice large objects and avoid the cost of slicing on small objects. If –minimum-size is specified, conditional slicing is enabled and works as follows.

The plugin builds a object size cache in memory. The key is the URL of the object. Only large object URLs are written to the cache. The object size cache uses CLOCK eviction algorithm in order to have lazy promotion behavior.

When a URL not found in the object size cache, the plugin treats the object as a small object. It will not intercept the request. The request is processed by ATS without any slice logic. Upon receiving a response, the slice plugin will check the response content length to update the object size cache if necessary.

When a large URL is requested for the first time, conditional slicing will not intercept that request since the URL is not known to be large. This will cause an ATS cache miss and the request will go to origin server. Slice plugin will turn off writing to cache for this response, because it expects to slice this object in future requests.

If the object size cache evicts a URL, the size of the object for that URL will need to be learned again in a subsequent request, and the behavior above will happen again.

If the URL is found in the object size cache, conditional slicing treats the object as a large object and will activate the slicing logic as described in the rest of this document.

If the client sends a range request, and that URL is not in the object size cache, the slice plugin will forward the range request to ATS core. It also attaches an internal header in order to deactivate the cache_range_requests plugin for this range request.

Important Notes

This plugin assumes that the content requested is cacheable.

Any first block server response that is not a 206 is passed directly down to the client. Any 200 responses are passed back through to the client.

Only the first server response block is used to evaluate any “If-” conditional headers. Subsequent server slice block requests remove these headers.

The only 416 response that this plugin handles itself is if the requested range is inside the last slice block but past the end of the asset contents. Other 416 responses are handled by the parent.

If a client aborts mid transaction the current slice block continues to be read from the server until it is complete to ensure that the block is cached.

Slice always makes blockbytes sized requests which are handled by cache_range_requests. The parent will trim those requests to account for the asset Content-Length so only the appropriate number of bytes are actually transferred and cached.

Effective URL remap

By default the plugin restores the Pristine Url which reuses the same remap rule for each slice block. This is wasteful in that it reruns the previous remap rules, and those remap rules must be smart enough to check for the existence of any headers they may have created the first time they have were visited.

To get around this the ‘–remap-host=<host>’ or ‘-r <host>’ option may be used. This requires an intermediate loopback remap to be defined which handles each slice block request.

This works well with any remap rules that use the url_sig or uri_signing plugins. As the client remap rule is not caching any plugins that manipulate the cache key would need to go into the loopback to parent remap rule.

NOTE: Requests NOT handled by the slice plugin (ie: HEAD requests) are handled as with a typical remap rule. GET requests intercepted by the slice plugin are virtually reissued into ATS and are proxied through another remap rule which must contain the cache_range_requests plugin

Examples:

map http://ats/ http://parent/ @plugin=slice.so @pparam=--remap-host=loopback
map http://loopback/ http://parent/ @plugin=cache_range_requests.so

Alternatively:

map http://ats/ http://parent/ @plugin=slice.so @pparam=-r @pparam=loopback
map http://loopback/ http://parent/ @plugin=cache_range_requests.so

Current Limitations

Since the Slice plugin is written as an intercept handler it loses the ability to use normal state machine hooks and transaction states. This functionality is handled by using the cache_range_requests plugin to interact with ATS.