Libguestfs - Sep 2019 - Re: [PATCH v4 07/12] v2v: nbdkit: Add the readahead filter unconditionally if it is available.

On Fri, Sep 20, 2019 at 10:28:18AM +0100, Richard W.M. Jones
wrote:
>The readahead filter is a self-configuring filter that makes
>sequential reads faster when the plugin is slow (and all of the
>plugins we use here are always slow).
>
>I observed the behaviour of the readahead filter with our qcow2
>overlay when converting a guest from a vCenter source.  Even when
>doing random reads, qemu issues 64K reads which happen to also be the
>minimum request size of the readahead filter, so there is no extra
>overhead.  When doing the sequential copy the readahead filter
>performed better than qemu curl’s readahead because it scaled the
>prefetched data appropriately on long contiguous stretches and then
>shrunk it back to 64K around fragmented parts of the base image.
>---
> v2v/nbdkit.ml | 19 ++++++++++++++++++-
> 1 file changed, 18 insertions(+), 1 deletion(-)
>
>diff --git a/v2v/nbdkit.ml b/v2v/nbdkit.ml
>index 4bbb8f043..b2d77f963 100644
>--- a/v2v/nbdkit.ml
>+++ b/v2v/nbdkit.ml
>@@ -49,6 +49,9 @@ type t = {
>
>   (* nbdkit plugin_name --dump-plugin output. *)
>   dump_plugin : (string * string) list;
>+
>+  (* nbdkit directory containing the filters. *)
>+  filterdir : string;
> }
>
> (* Check that nbdkit is available and new enough. *)
>@@ -105,6 +108,12 @@ let common_create plugin_name plugin_args plugin_env
>   error_unless_nbdkit_min_version dump_config;
>   error_unless_nbdkit_compiled_with_selinux dump_config;
>
>+  (* Get the filterdir. *)
>+  let filterdir >+    try List.assoc "filterdir" dump_config
>+    with Not_found ->
>+      error (f_"nbdkit --dump-config output did not contain
filterdir") in
>+
>   (* Get the nbdkit plugin_name --dump-plugin output, which also
>    * checks that the plugin is available and loadable.
>    *)
>@@ -133,9 +142,17 @@ let common_create plugin_name plugin_args plugin_env
>   if have_selinux then (        (* label the socket so qemu can open it *)
>     add_arg "--selinux-label"; add_arg
"system_u:object_r:svirt_socket_t:s0"
>   );
>+
>+  (* Adding the readahead filter is always a win for our access
>+   * patterns.  However if it doesn't exist don't worry.
>+   *)
>+  if Sys.file_exists (filterdir // "nbdkit-readahead-filter.so")
then (
>+    add_arg "--filter"; add_arg "readahead"
Just out of curiosity, wouldn't it be cleaner, at least in the future, to
add an
option in nbdkit to report filters it can load?  Of course it would require
newer nbdkit in the long run.
>+  );
>+
>   let args = get_args () @ [ plugin_name ] @ plugin_args in
>
>-  { plugin_name; args; env; dump_config; dump_plugin }
>+  { plugin_name; args; env; dump_config; dump_plugin; filterdir }
>
> (* VDDK libraries are located under lib32/ or lib64/ relative to the
>  * libdir.  Note this is unrelated to Linux multilib or multiarch.
>-- 
>2.23.0
>
>_______________________________________________
>Libguestfs mailing list
>Libguestfs@redhat.com
>https://www.redhat.com/mailman/listinfo/libguestfs

On 9/20/19 9:04 AM, Martin Kletzander wrote:
> On Fri, Sep 20, 2019 at 10:28:18AM +0100, Richard W.M. Jones wrote:
>> The readahead filter is a self-configuring filter that makes
>> sequential reads faster when the plugin is slow (and all of the
>> plugins we use here are always slow).
>>
>> @@ -133,9 +142,17 @@ let common_create plugin_name plugin_args
>> plugin_env >>   if have_selinux then (        (* label the socket
so qemu can open
>> it *)
>>     add_arg "--selinux-label"; add_arg
>> "system_u:object_r:svirt_socket_t:s0"
>>   );
>> +
>> +  (* Adding the readahead filter is always a win for our access
>> +   * patterns.  However if it doesn't exist don't worry.
>> +   *)
>> +  if Sys.file_exists (filterdir //
"nbdkit-readahead-filter.so") then (
>> +    add_arg "--filter"; add_arg "readahead"
> 
> Just out of curiosity, wouldn't it be cleaner, at least in the future,
> to add an
> option in nbdkit to report filters it can load?  Of course it would require
> newer nbdkit in the long run.
> 
Checking for file existence for filters is somewhat less fragile than
for plugins, because all filters are built in-tree (we've specifically
documented that we don't provide ABI guarantees for filters, so the only
sane way to use a filter is to compile it at the same time/version as
the nbdkit binary that will load it).  But you do have a point that
checking for files is still more fragile than just asking nbdkit whether
a given filter exists:

$ nbdkit --dump-plugin --filter=nosuch null
nbdkit: error: cannot open filter 'nosuch':
/usr/lib64/nbdkit/filters/nbdkit-nosuch-filter.so: cannot open shared
object file: No such file or directory
$ echo $?
1
$ nbdkit --dump-plugin --filter=cache null
path=/usr/lib64/nbdkit/plugins/nbdkit-null-plugin.so
name=null
...
$ echo $?
0

Similarly for --help.  But notice:

$ diff -u <(nbdkit --dump-plugin --filter=cache null) \
   <(nbdkit --dump-plugin null)
$ diff -u <(nbdkit --help --filter=cache null) <(nbdkit --help null)
--- /dev/fd/63	2019-09-20 09:31:43.184428823 -0500
+++ /dev/fd/62	2019-09-20 09:31:43.185428824 -0500
@@ -23,15 +23,6 @@

 Please read the nbdkit(1) manual page for full usage.

-filter: cache (nbdkit caching filter)
-(/usr/lib64/nbdkit/filters/nbdkit-cache-filter.so)
-cache=MODE                Set cache MODE, one of writeback (default),
-                          writethrough, or unsafe.
-cache-on-read=BOOL        Set to true to cache on reads (default false).
-cache-max-size=SIZE       Set maximum space used by cache.
-cache-high-threshold=PCT  Percentage of max size where reclaim begins.
-cache-low-threshold=PCT   Percentage of max size where reclaim ends.
-
 plugin: null
 (/usr/lib64/nbdkit/plugins/nbdkit-null-plugin.so)
 size=<SIZE>             Size of the backing disk


So we could probably improve --dump-plugin to ALSO show details about
the filter (such as a second path=), the way --help does, or even add a
'nbdkit --dump-filter filtername' to work in isolation.  But while
you'd
have to wait for a future nbdkit release to get more details from
--dump-plugin, and/or a way to list all available filters (think search
permissions on a directory), you already have a way with current nbdkit
to check existence of a filter (think read permissions on a directory).

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org

Also of note in this conversation:

https://github.com/libguestfs/nbdkit/blob/master/docs/nbdkit-probing.pod

which currently doesn't mention filters, but ought to once we settle
on the definitive way to detect them.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-df lists disk usage of guests without needing to install any
software inside the virtual machine.  Supports Linux and Windows.
http://people.redhat.com/~rjones/virt-df/

On Fri, Sep 20, 2019 at 09:33:06AM -0500, Eric Blake
wrote:
>On 9/20/19 9:04 AM, Martin Kletzander wrote:
>> On Fri, Sep 20, 2019 at 10:28:18AM +0100, Richard W.M. Jones wrote:
>>> The readahead filter is a self-configuring filter that makes
>>> sequential reads faster when the plugin is slow (and all of the
>>> plugins we use here are always slow).
>>>
>
>>> @@ -133,9 +142,17 @@ let common_create plugin_name plugin_args
>>> plugin_env >>>   if have_selinux then (        (* label
the socket so qemu can open
>>> it *)
>>>     add_arg "--selinux-label"; add_arg
>>> "system_u:object_r:svirt_socket_t:s0"
>>>   );
>>> +
>>> +  (* Adding the readahead filter is always a win for our access
>>> +   * patterns.  However if it doesn't exist don't worry.
>>> +   *)
>>> +  if Sys.file_exists (filterdir //
"nbdkit-readahead-filter.so") then (
>>> +    add_arg "--filter"; add_arg "readahead"
>>
>> Just out of curiosity, wouldn't it be cleaner, at least in the
future,
>> to add an
>> option in nbdkit to report filters it can load?  Of course it would
require
>> newer nbdkit in the long run.
>>
>
>Checking for file existence for filters is somewhat less fragile than
>for plugins, because all filters are built in-tree (we've specifically
>documented that we don't provide ABI guarantees for filters, so the only
>sane way to use a filter is to compile it at the same time/version as
>the nbdkit binary that will load it).  But you do have a point that
>checking for files is still more fragile than just asking nbdkit whether
>a given filter exists:
>
>$ nbdkit --dump-plugin --filter=nosuch null
Oh, this is way better than my:

  nbdkit --filter=nosuch memory size=1M --run true
>nbdkit: error: cannot open filter 'nosuch':
>/usr/lib64/nbdkit/filters/nbdkit-nosuch-filter.so: cannot open shared
>object file: No such file or directory
>$ echo $?
>1
>$ nbdkit --dump-plugin --filter=cache null
>path=/usr/lib64/nbdkit/plugins/nbdkit-null-plugin.so
>name=null
>...
>$ echo $?
>0
>
>Similarly for --help.  But notice:
>
>$ diff -u <(nbdkit --dump-plugin --filter=cache null) \
>   <(nbdkit --dump-plugin null)
>$ diff -u <(nbdkit --help --filter=cache null) <(nbdkit --help null)
>--- /dev/fd/63	2019-09-20 09:31:43.184428823 -0500
>+++ /dev/fd/62	2019-09-20 09:31:43.185428824 -0500
>@@ -23,15 +23,6 @@
>
> Please read the nbdkit(1) manual page for full usage.
>
>-filter: cache (nbdkit caching filter)
>-(/usr/lib64/nbdkit/filters/nbdkit-cache-filter.so)
>-cache=MODE                Set cache MODE, one of writeback (default),
>-                          writethrough, or unsafe.
>-cache-on-read=BOOL        Set to true to cache on reads (default false).
>-cache-max-size=SIZE       Set maximum space used by cache.
>-cache-high-threshold=PCT  Percentage of max size where reclaim begins.
>-cache-low-threshold=PCT   Percentage of max size where reclaim ends.
>-
But this is still better because it also allows checking for the filter
parameters had they been changed.

In this case I feel like there is no need to add anything else, I did not know
that this already exists.  I don't even know if this is really needed, I was
just curious if it would be nicer.
> plugin: null
> (/usr/lib64/nbdkit/plugins/nbdkit-null-plugin.so)
> size=<SIZE>             Size of the backing disk
>
>
>So we could probably improve --dump-plugin to ALSO show details about
>the filter (such as a second path=), the way --help does, or even add a
>'nbdkit --dump-filter filtername' to work in isolation.  But while
you'd
>have to wait for a future nbdkit release to get more details from
>--dump-plugin, and/or a way to list all available filters (think search
>permissions on a directory), you already have a way with current nbdkit
>to check existence of a filter (think read permissions on a directory).
>
>--
>Eric Blake, Principal Software Engineer
>Red Hat, Inc.           +1-919-301-3226
>Virtualization:  qemu.org | libvirt.org
>

On Fri, Sep 20, 2019 at 09:33:06AM -0500, Eric Blake
wrote:
> Checking for file existence for filters is somewhat less fragile than
> for plugins, because all filters are built in-tree (we've specifically
> documented that we don't provide ABI guarantees for filters, so the
only
> sane way to use a filter is to compile it at the same time/version as
> the nbdkit binary that will load it).  But you do have a point that
> checking for files is still more fragile than just asking nbdkit whether
> a given filter exists:
> 
> $ nbdkit --dump-plugin --filter=nosuch null
> nbdkit: error: cannot open filter 'nosuch':
> /usr/lib64/nbdkit/filters/nbdkit-nosuch-filter.so: cannot open shared
> object file: No such file or directory
This is definitely the best we can do now, and I've posted a patch
suggesting this change for the nbdkit-probing(1) man page.

However it's not without a subtle problem: It requires the null plugin
to be present.  It's possible to ship nbdkit on its own.  On Fedora
try installing just the nbdkit-server package.  The null plugin won't
be available so this command will always fail.

So we do also need to change nbdkit to allow easier probing for
plugins or filters in the long term.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html