4:32 p.m.
On 11/16/2017 10:19 AM, Daniel P. Berrange wrote:
While we have collective knowledge about the support status of
various
parts of libvirt, this has never been formally documented, leaving our
users to guess.
Signed-off-by: Daniel P. Berrange <berrange(a)redhat.com>
---
docs/docs.html.in | 3 +
docs/support.html.in | 257 +++++++++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 260 insertions(+)
create mode 100644 docs/support.html.in
diff --git a/docs/docs.html.in b/docs/docs.html.in
index 931da85424..a8d544f83f 100644
--- a/docs/docs.html.in
+++ b/docs/docs.html.in
@@ -110,6 +110,9 @@
<dt><a href="drivers.html">Drivers</a></dt>
<dd>Hypervisor specific driver information</dd>
+ <dt><a href="support.html">Support
guarantees</a></dt>
+ <dd>Details of support status for various interfaces</dd>
+
<dt><a href="hvsupport.html">Driver
support</a></dt>
<dd>matrix of API support per hypervisor per release</dd>
diff --git a/docs/support.html.in b/docs/support.html.in
new file mode 100644
index 0000000000..6f34a35eb7
--- /dev/null
+++ b/docs/support.html.in
@@ -0,0 +1,257 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml">
+ <body>
+ <h1>Support guarantees</h1>
+
+ <ul id="toc"></ul>
+
+ <p>
+ This document will outline the support status / guarantees around the
+ very interfaces that libvirt exposes to applications and/or system
+ adminstrators. The intent is to help users understand what features they
administrators
+ can rely upon in particular scenarios, and whether they are
likely to
+ suffer disruption during upgrades
+ </p>
+
+ <h2><a id="publicAPI">Primary public API</a></h2>
+
+ <p>
+ The main public API provided by <code>libvirt.so</code> and described
+ in <code>libvirt/libvirt.h</code> exposes the primary hypervisor
+ agnostic management interface of libvirt. This API has the strongest
+ guarantee of any part of libvirt with a promise to keep backwards
+ compatibility forever. Specific details are as follows:
+ </p>
+
+ <dl>
+ <dt>Functions</dt>
+ <dd>Functions will never be removed from the public API, and will
+ never have parameters added, removed or changed in their signature.
+ IOW they will be ABI compatible forever. The semantics implied by
+ a specific set of parameters passed to the function will remain
+ unchanged. Where a parameter accepts a bitset of feature flags, or
+ an enumerated value, further flags / enum values may be supported
+ in the future. Where a parameter accepts one of a set of related
+ constants, further constants may be supported in the future.
+ </dd>
+ <dt>Struct types</dt>
+ <dd>Once defined in a release, struct definitions will never have any
+ fields add, removed or changed in any way. Their size and layout is
+ fixed forever. If a struct name starts with an underscore, it is
+ considered acceptable to rename it. Applications should thus always
+ use the corresponding typedef in preference to the struct name.
+ </dd>
+ <dt>Union types</dt>
+ <dd>Once defined in a release, union definitions will never have any
+ existing fields removed or changed. New union choices may be added,
+ provided that they don't change the size of the existing union
+ definition. If a struct name starts with an underscore, it is
+ considered acceptable to rename it. Applications should thus always
+ use the corresponding typedef in preference to the struct name.
+ </dd>
+ <dt>Type definitions</dt>
+ <dd>Most custom data types used in the APIs have corresponding typedefs
+ provided for their stable names. The typedefs should always be used
+ in preference to the underlying data type name, as the latter are not
+ guaranteed to be stable.
+ </dd>
+ <dt>Enumerations</dt>
+ <dd>Once defined in a release, existing enumeration values will never
+ be removed or renamed. New enumeration values may be introduced at
+ any time. Every enumeration will have a '_LAST' value which indicates
+ the current highest enumeration value, which may increase with new
+ releases. If an enumeration name starts with an underscore, it is
+ considered acceptable to rename it. Applications should thus always
+ use the corresponding typedef in preference to the enum name.
+ </dd>
+ <dt>Constants</dt>
+ <dd>Once defined in a release, existing constants will never be removed
+ or have their value changed. Most constants are grouped into related
+ sets, and within each set, new constants may be introduced. APIs which
+ use the constants may thus accept or return new constant values over
+ time.
+ </dd>
+ <dt>Symbol versions</dt>
+ <dd>Where the platform library format permits, APIs defined in libvirt.so
+ library will have version information associated. Each API will be
+ tagged with the version in which it was introduced, and this won't
+ be changed thereafter.
+ </dd>
+ </dl>
+
+ <h2><a id="hvAPI">Hypervisor specific
APIs</a></h2>
+
+ <p>
+ A number of hypervisor drivers provide additional libraries with hypervisor
+ specific APIs, extending the core libvirt API. These add-on libraries follow
+ the same general principles described above, however, they are
<strong>not</strong>
+ not guranteed to be preserved forever. The project reserves the right to remove
extra 'not'
+ hypervisor specific APIs in any new release, or to change
their semantics.
+ That said the project will endeavour to maintain API compatibility for as long
+ as is practical.
+ </p>
+
+ <p>
+ Use of some hypervisor specific APIs may result in the running guest being
+ marked as "tainted" if the API is at risk of having unexpected
interactions
+ with normal libvirt operations. An application which chooses to make use of
+ hypervisor specific APIs should validate their operation with each new release
+ of libvirt and each new release of the underlying hypervisor. The semantics
+ may change in unexpected ways, or have unforseen interactions with libvirt's
unforeseen
+ operation.
+ </p>
+
+ <h2><a id="apierrors">Error reporting</a></h2>
+
+ <p>
+ Almost API calls are subject to failure and so will report error codes and
s/Almost/Most/
+ messages. Libvirt defines error codes for a wide variety of
scenarios, some
+ are very targetted to specific problems, while others are general purpose
targeted ? Or is this like tunnelled vs tunneled :-).
+ for broad classes of problem. Over time the error codes
reported are liable
+ to, usually changing from a generic error to a more specific error. Thus
s/liable to,/liable to change,/
+ applications should be careful about checking for &
taking action upon
+ specific error codes, as the behaviour is liable to change across releases.
+ </p>
+
+ <h2><a id="xmlschema">XML schemas</a></h2>
+
+ <p>
+ The main objects exposed via the primary libvirt public API are usually
+ configured via XML documents following specific schemas. The XML schemas
+ are considered to be stable formats, whose compatibility will be maintained
+ forever. Specific details are as follows:
+ </p>
+
+ <dl>
+ <dt>Attributes</dt>
+ <dd>Attributes defined on an XML element will never be removed or
+ renamed. New attributes may be defined. If the set of valid values
+ for an attribute are determined by an enumeration, the permitted
+ values will never be removed or renamed, only new values defined.
+ None the less, specific hypervisors may reject usage of certain
+ values according to their feature set.</dd>
+ <dt>Elements</dt>
+ <dd>Elements defined will never be removed or renamed. New child
+ elements may be defined at any time. In places where only a
+ single instance of a named XML element is used, future versions
+ may be extended to permit multiple instances of the named XML
+ element to be used. An element which currently has no content
+ may later gain child elements.
+ </dd>
+ </dl>
+
+ <p>
+ Some hypervisor drivers may choose to allow use of hypervisor specific
+ extensions to the XML documents. These extensions will always be
+ contained within a hypervisor specific XML namespace. There is generally
+ no guarantee of long term support for the hypervisor specific extensions
+ across releases, though the project will endeavour to preserve them as
+ long as is possible. Applications choosing to use hypervisor specific
+ extensions should validate their operation against new libvirt or
+ hypervisor releases.
+ </p>
+
+ <h2><a id="configfiles">Configuration
files</a></h2>
+
+ <p>
+ A number of programs / daemons provided libvirt rely on host filesystem
+ configuration files. These configuration files are accompanied by augeas
+ lens for easy manipulation by applications. There is in general no
+ guarantee that parameters available in the configuration file will be
+ preserved across releases, though the project will endeavour to preserve
+ them as long as is possible. If a configuration option is dropped from
+ the file, the augeas lens will retain the ability to read that configuration
+ parameter, so that it is able to read & update historically modified
+ files.
+
+ The default configuration files ship with all parameters commented out
+ such that a deployment relies on the built-in defaults of the application
+ in question. There is no guarantee that the defaults will remain the same
+ across releases. An deployments that expects a particular value for a
s/An deployments/A deployment/
+ configuration parameter should consider defining it
explicitly, instead
+ of relying on the defaults.
+ </p>
+
+ <h2><a id="hvdrivers">Hypervisor drivers</a></h2>
+
+ <p>
+ The libvirt project provides support for a wide variety of hypervisor
+ drivers. These drivers target certain versions of the hypervisor's
+ underlying management APIs. In general libvirt aims to work with any
+ hypervisor version that is still broadly supported by its vendor.
+ When a vendor discontinues support for a particular hypervisor
+ version it will be dropped by libvirt. Libvirt may choose to drop
+ support for a particular hypervisor version prior to the vendor
+ ending support, if it deems that the likely usage is too small to
+ justify the ongoing maintenance cost.
+ </p>
+ <p>
+ Each hypervisor release will implement a distinct subset of features
+ that can be expressed in the libvirt APIs and XML formats. While the
+ XML schema syntax will be stable across releases, libvirt is unable
+ to promise that it will always be able to support usage of the same
+ features across hypervisor releases. Where a hypervisor changes the
+ way a feature is implemented, the project will endeavour to adapt
+ to the new implementation to provide the same semantics. In cases
+ where the feature is discontinued by the hypervisor, libvirt will
+ return an error indicating it is no supported. Likewise libvirt will
s/no/not/
+ make reasonable efforts to keep API calls working across
hypervisor
+ releases even if the underlying implementation changes. In cases where
+ this is impossible, an suitable error will be reported. The list of
+ APIs which have implementations <a href="hvsupport.html">is
detailed separately</a>.
+ </p>
+
+ <h2><a id="rpcproto">RPC protocol</a></h2>
+
+ <p>
+ For some hypervisor drivers, the libvirt.so library communicates with
+ separate libvirt daemons to perform work. This communication takes
+ place over a binary RPC protocol defined by libvirt. The protocol uses
+ the XDR format for data encoding, and the message packet format is
+ defined in libvirt source code.
+ </p>
+ <p>
+ Applications are encouraged to use the primary libvirt.so library which
+ transparently talks to the daemons, so that they are not exposed to the
+ hypervisor driver specific details. None the less, the RPC protocol
+ associated with the libvirtd is considered to be a long term stable ABI.
+ It will only ever have new messages added to it, existing messages will
+ not be removed, nor have their contents changed. Thus if an application
+ does wish to provide its own client side implementation of the RPC
+ protocol this is supported, with the caveat that the application will
+ loose the ability to work with certain hypervisors libvirt supports.
+ The project reserves the right to define new authentication and encryption
+ options for the protocol, and the defaults used in this area may change
+ over time. This is particularly true of the TLS ciphers permitted. Thus
+ applications choosing to implement the RPC protocol must be prepared to
+ track support for new security options. If defaults are changed, however,
+ it will generally be possible to reconfigure the daemon to use the old
+ defaults, albeit with possible implications for system security.
+ </p>
+
+ <p>
+ Other daemons besides, libvirtd, also use the same RPC protocol, but
+ with different message types defined. These RPC protocols are all
+ considered to be private implementation that are liable to change
s/implementation/implementations/ ?
+ at any time. Applications must not attempt to talk to these
other
+ daemons directly.
+ </p>
+
+ <h2><a id="virsh">virsh client</a></h2>
+
+ <p>
+ The virsh program provides a simple client to interact with an arbitrary libvirt
+ hypevisor connection. Since it uses the primary public API of libvirt, it should
+ generally inherit the guarantees associated with that API, and with the
hypervisor
+ driver. The commands that virsh exposes, and the arguments they accept are all
+ considered to be long term stable. Existing commands and arguments will not be
+ removed or renamed. New commands and arguments may be added in new releases.
+ The text output format produced by virsh commands is not generally guaranteed to
+ be stable if it contains compound data (eg formatted tables or lists). Commands
+ which output single data items (ie an object name, or an XML document), can be
+ treated as having stable format.
+ </p>
+
+ </body>
+</html>
Reviewed-by: Jim Fehlig <jfehlig(a)suse.com>
Regards,
Jim