| Message ID | 20210730122643.2043-3-jgross@suse.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | tools/xenstore: set resource limits of xenstored | expand |
Juergen Gross writes ("[PATCH v3 2/2] tools/xenstore: set open file descriptor limit for xenstored"):
> Add a configuration item for the maximum number of domains xenstored
> should support and set the limit of open file descriptors accordingly.
>
> For HVM domains there are up to 5 socket connections per domain (2 by
> the xl daemon process, and 3 by qemu). So set the ulimit for xenstored
> to 5 * XENSTORED_MAX_DOMAINS + 100 (the "+ 100" is for some headroom,
> like logging, event channel device, etc.).
...
> +## Type: integer
> +## Default: 32768
> +#
> +# Select maximum number of domains supported by xenstored.
> +# Only evaluated if XENSTORETYPE is "daemon".
> +#XENSTORED_MAX_N_DOMAINS=32768
I approve of doing something about the fd limit. I have some qualms
about the documentation.
The documentation doesn't say what happens if this limit is exceeded.
Also the default of 32758 suggests that we actually support that many
domains. I don't think we do...
I didn't find anything in SUPPORT.md about how many guests we support
but I wouldn't want this setting here to imply full support for 32768
domains.
If you don't want to tackle this can of works, maybe add this:
# This just controls some resource limits for xenstored; if the
# limit is exceeded, xenstored will stop being able to function
# properly for additional guests. The default value is so large
# that it won't be exceeded in a supported configuration, but
# should not be taken to mean that the whole Xen system is
# guaranteed to work properly with that many guests.
Julien, did you ask for this to be made configurable ? Having written
the text above, I wonder if it wouldn't just be better to
unconditionally set it to "unlimited" rather than offering footgun
dressed up like a tuneable...
If xenstored does go mad or leadk lots of fds, things are basically
stuffed in anycase. Having its syscalls fail with EMFILE is not
really going to help.
Ian.
Hi Ian, On 30/07/2021 14:35, Ian Jackson wrote: > Juergen Gross writes ("[PATCH v3 2/2] tools/xenstore: set open file descriptor limit for xenstored"): >> Add a configuration item for the maximum number of domains xenstored >> should support and set the limit of open file descriptors accordingly. >> >> For HVM domains there are up to 5 socket connections per domain (2 by >> the xl daemon process, and 3 by qemu). So set the ulimit for xenstored >> to 5 * XENSTORED_MAX_DOMAINS + 100 (the "+ 100" is for some headroom, >> like logging, event channel device, etc.). > ... >> +## Type: integer >> +## Default: 32768 >> +# >> +# Select maximum number of domains supported by xenstored. >> +# Only evaluated if XENSTORETYPE is "daemon". >> +#XENSTORED_MAX_N_DOMAINS=32768 > > I approve of doing something about the fd limit. I have some qualms > about the documentation. > > The documentation doesn't say what happens if this limit is exceeded. > Also the default of 32758 suggests that we actually support that many > domains. I don't think we do... > > I didn't find anything in SUPPORT.md about how many guests we support > but I wouldn't want this setting here to imply full support for 32768 > domains. > > If you don't want to tackle this can of works, maybe add this: > > # This just controls some resource limits for xenstored; if the > # limit is exceeded, xenstored will stop being able to function > # properly for additional guests. The default value is so large > # that it won't be exceeded in a supported configuration, but > # should not be taken to mean that the whole Xen system is > # guaranteed to work properly with that many guests. > > Julien, did you ask for this to be made configurable ? Having written > the text above, I wonder if it wouldn't just be better to > unconditionally set it to "unlimited" rather than offering footgun > dressed up like a tuneable... So in v1 (see [1]), Juergen wanted to raise the limit. I assumed this meant that the default limit (configured by the system may not be enough). I felt this was wrong to impose an higher limit on everyone when an admin may know the maximum number of domains. By "unlimited", do you mean the calling "ulimit" (or whatever is used for configuring FDs) with unlimited? If so, I would be OK with that. My main was was to move the raising the limit outside Xenstored because: 1) This is easier for an admin to tweak it (in particular the OOM) 2) It feels wrong to me that the daemon chose the limits 3) An admin can enforce it Cheers, [1] 1e38cce0-6960-ac21-b349-dac8551e23ed@xen.org
On 30.07.21 19:14, Julien Grall wrote: > Hi Ian, > > On 30/07/2021 14:35, Ian Jackson wrote: >> Juergen Gross writes ("[PATCH v3 2/2] tools/xenstore: set open file >> descriptor limit for xenstored"): >>> Add a configuration item for the maximum number of domains xenstored >>> should support and set the limit of open file descriptors accordingly. >>> >>> For HVM domains there are up to 5 socket connections per domain (2 by >>> the xl daemon process, and 3 by qemu). So set the ulimit for xenstored >>> to 5 * XENSTORED_MAX_DOMAINS + 100 (the "+ 100" is for some headroom, >>> like logging, event channel device, etc.). >> ... >>> +## Type: integer >>> +## Default: 32768 >>> +# >>> +# Select maximum number of domains supported by xenstored. >>> +# Only evaluated if XENSTORETYPE is "daemon". >>> +#XENSTORED_MAX_N_DOMAINS=32768 >> >> I approve of doing something about the fd limit. I have some qualms >> about the documentation. >> >> The documentation doesn't say what happens if this limit is exceeded. >> Also the default of 32758 suggests that we actually support that many >> domains. I don't think we do... >> >> I didn't find anything in SUPPORT.md about how many guests we support >> but I wouldn't want this setting here to imply full support for 32768 >> domains. >> >> If you don't want to tackle this can of works, maybe add this: >> >> # This just controls some resource limits for xenstored; if the >> # limit is exceeded, xenstored will stop being able to function >> # properly for additional guests. The default value is so large >> # that it won't be exceeded in a supported configuration, but >> # should not be taken to mean that the whole Xen system is >> # guaranteed to work properly with that many guests. >> >> Julien, did you ask for this to be made configurable ? Having written >> the text above, I wonder if it wouldn't just be better to >> unconditionally set it to "unlimited" rather than offering footgun >> dressed up like a tuneable... > > So in v1 (see [1]), Juergen wanted to raise the limit. I assumed this > meant that the default limit (configured by the system may not be enough). > > I felt this was wrong to impose an higher limit on everyone when an > admin may know the maximum number of domains. > > By "unlimited", do you mean the calling "ulimit" (or whatever is used > for configuring FDs) with unlimited? > > If so, I would be OK with that. My main was was to move the raising the > limit outside Xenstored because: > 1) This is easier for an admin to tweak it (in particular the OOM) > 2) It feels wrong to me that the daemon chose the limits > 3) An admin can enforce it Coming back to this series, I'm puzzled now. Julien, you didn't want me to raise the limit to a specific number covering the maximum possible number of domains, because you thought this might result in xenstored hogging huge numbers of file descriptors in case of a bug. Why is unlimited better then? This will make the possible number even larger. I'd really like to know how to come to an acceptable end result soon. Right now the max number of domains supported by xenstored is just limited by an arbitrary system wide limit. Juergen
Hi Juergen, On 31/08/2021 13:11, Juergen Gross wrote: > On 30.07.21 19:14, Julien Grall wrote: >> Hi Ian, >> >> On 30/07/2021 14:35, Ian Jackson wrote: >>> Juergen Gross writes ("[PATCH v3 2/2] tools/xenstore: set open file >>> descriptor limit for xenstored"): >>>> Add a configuration item for the maximum number of domains xenstored >>>> should support and set the limit of open file descriptors accordingly. >>>> >>>> For HVM domains there are up to 5 socket connections per domain (2 by >>>> the xl daemon process, and 3 by qemu). So set the ulimit for xenstored >>>> to 5 * XENSTORED_MAX_DOMAINS + 100 (the "+ 100" is for some headroom, >>>> like logging, event channel device, etc.). >>> ... >>>> +## Type: integer >>>> +## Default: 32768 >>>> +# >>>> +# Select maximum number of domains supported by xenstored. >>>> +# Only evaluated if XENSTORETYPE is "daemon". >>>> +#XENSTORED_MAX_N_DOMAINS=32768 >>> >>> I approve of doing something about the fd limit. I have some qualms >>> about the documentation. >>> >>> The documentation doesn't say what happens if this limit is exceeded. >>> Also the default of 32758 suggests that we actually support that many >>> domains. I don't think we do... >>> >>> I didn't find anything in SUPPORT.md about how many guests we support >>> but I wouldn't want this setting here to imply full support for 32768 >>> domains. >>> >>> If you don't want to tackle this can of works, maybe add this: >>> >>> # This just controls some resource limits for xenstored; if the >>> # limit is exceeded, xenstored will stop being able to function >>> # properly for additional guests. The default value is so large >>> # that it won't be exceeded in a supported configuration, but >>> # should not be taken to mean that the whole Xen system is >>> # guaranteed to work properly with that many guests. >>> >>> Julien, did you ask for this to be made configurable ? Having written >>> the text above, I wonder if it wouldn't just be better to >>> unconditionally set it to "unlimited" rather than offering footgun >>> dressed up like a tuneable... >> >> So in v1 (see [1]), Juergen wanted to raise the limit. I assumed this >> meant that the default limit (configured by the system may not be >> enough). >> >> I felt this was wrong to impose an higher limit on everyone when an >> admin may know the maximum number of domains. >> >> By "unlimited", do you mean the calling "ulimit" (or whatever is used >> for configuring FDs) with unlimited? >> >> If so, I would be OK with that. My main was was to move the raising >> the limit outside Xenstored because: >> 1) This is easier for an admin to tweak it (in particular the OOM) >> 2) It feels wrong to me that the daemon chose the limits >> 3) An admin can enforce it > > Coming back to this series, I'm puzzled now. > > Julien, you didn't want me to raise the limit to a specific number > covering the maximum possible number of domains, because you thought > this might result in xenstored hogging huge numbers of file descriptors > in case of a bug. Why is unlimited better then? This will make the > possible number even larger. I don't think I suggested the unlimited number is better... My main objection in your original approach is you set an arbitrary limit you in Xenstored (which may not apply at all) and don't offer a way to the admin to tweak it. If the limit is set outside of Xenstored, then it becomes much easier for someone to just tweak the init script. I don't have a strong opinion on whether the default limit should be "unlimited" or a fixed number. Cheers,
On 31/08/2021 13:30, Julien Grall wrote: > Hi Juergen, > > On 31/08/2021 13:11, Juergen Gross wrote: >> On 30.07.21 19:14, Julien Grall wrote: >>> Hi Ian, >>> >>> On 30/07/2021 14:35, Ian Jackson wrote: >>>> Juergen Gross writes ("[PATCH v3 2/2] tools/xenstore: set open file >>>> descriptor limit for xenstored"): >>>>> Add a configuration item for the maximum number of domains xenstored >>>>> should support and set the limit of open file descriptors >>>>> accordingly. >>>>> >>>>> For HVM domains there are up to 5 socket connections per domain (2 by >>>>> the xl daemon process, and 3 by qemu). So set the ulimit for >>>>> xenstored >>>>> to 5 * XENSTORED_MAX_DOMAINS + 100 (the "+ 100" is for some headroom, >>>>> like logging, event channel device, etc.). >>>> ... >>>>> +## Type: integer >>>>> +## Default: 32768 >>>>> +# >>>>> +# Select maximum number of domains supported by xenstored. >>>>> +# Only evaluated if XENSTORETYPE is "daemon". >>>>> +#XENSTORED_MAX_N_DOMAINS=32768 >>>> >>>> I approve of doing something about the fd limit. I have some qualms >>>> about the documentation. >>>> >>>> The documentation doesn't say what happens if this limit is exceeded. >>>> Also the default of 32758 suggests that we actually support that many >>>> domains. I don't think we do... >>>> >>>> I didn't find anything in SUPPORT.md about how many guests we support >>>> but I wouldn't want this setting here to imply full support for 32768 >>>> domains. >>>> >>>> If you don't want to tackle this can of works, maybe add this: >>>> >>>> # This just controls some resource limits for xenstored; if the >>>> # limit is exceeded, xenstored will stop being able to function >>>> # properly for additional guests. The default value is so large >>>> # that it won't be exceeded in a supported configuration, but >>>> # should not be taken to mean that the whole Xen system is >>>> # guaranteed to work properly with that many guests. >>>> >>>> Julien, did you ask for this to be made configurable ? Having written >>>> the text above, I wonder if it wouldn't just be better to >>>> unconditionally set it to "unlimited" rather than offering footgun >>>> dressed up like a tuneable... >>> >>> So in v1 (see [1]), Juergen wanted to raise the limit. I assumed >>> this meant that the default limit (configured by the system may not >>> be enough). >>> >>> I felt this was wrong to impose an higher limit on everyone when an >>> admin may know the maximum number of domains. >>> >>> By "unlimited", do you mean the calling "ulimit" (or whatever is >>> used for configuring FDs) with unlimited? >>> >>> If so, I would be OK with that. My main was was to move the raising >>> the limit outside Xenstored because: >>> 1) This is easier for an admin to tweak it (in particular the OOM) >>> 2) It feels wrong to me that the daemon chose the limits >>> 3) An admin can enforce it >> >> Coming back to this series, I'm puzzled now. >> >> Julien, you didn't want me to raise the limit to a specific number >> covering the maximum possible number of domains, because you thought >> this might result in xenstored hogging huge numbers of file descriptors >> in case of a bug. Why is unlimited better then? This will make the >> possible number even larger. > > I don't think I suggested the unlimited number is better... My main > objection in your original approach is you set an arbitrary limit you > in Xenstored (which may not apply at all) and don't offer a way to the > admin to tweak it. > > If the limit is set outside of Xenstored, then it becomes much easier > for someone to just tweak the init script. I don't have a strong > opinion on whether the default limit should be "unlimited" or a fixed > number. xenstored is TCB. It needs a large number of FDs, and can be trusted with unlimited. Also, like xenconsoled, we can calculate an upper bound, which is derived from the ABI limit of 32k domids. All you're haggling over is the error semantics in the case of: 1) the upper bound calculation is wrong, or 2) there is an fd leak Personally, I think a fixed calculation is right, so fd leaks can be spotted more obviously. An admin knob is not helpful - higher than the upper bound is just wasteful, while lower will cause malfunctions. ~Andrew
Hi Andrew, On 31/08/2021 13:37, Andrew Cooper wrote: > On 31/08/2021 13:30, Julien Grall wrote: >> Hi Juergen, >> >> On 31/08/2021 13:11, Juergen Gross wrote: >>> On 30.07.21 19:14, Julien Grall wrote: >>>> Hi Ian, >>>> >>>> On 30/07/2021 14:35, Ian Jackson wrote: >>>>> Juergen Gross writes ("[PATCH v3 2/2] tools/xenstore: set open file >>>>> descriptor limit for xenstored"): >>>>>> Add a configuration item for the maximum number of domains xenstored >>>>>> should support and set the limit of open file descriptors >>>>>> accordingly. >>>>>> >>>>>> For HVM domains there are up to 5 socket connections per domain (2 by >>>>>> the xl daemon process, and 3 by qemu). So set the ulimit for >>>>>> xenstored >>>>>> to 5 * XENSTORED_MAX_DOMAINS + 100 (the "+ 100" is for some headroom, >>>>>> like logging, event channel device, etc.). >>>>> ... >>>>>> +## Type: integer >>>>>> +## Default: 32768 >>>>>> +# >>>>>> +# Select maximum number of domains supported by xenstored. >>>>>> +# Only evaluated if XENSTORETYPE is "daemon". >>>>>> +#XENSTORED_MAX_N_DOMAINS=32768 >>>>> >>>>> I approve of doing something about the fd limit. I have some qualms >>>>> about the documentation. >>>>> >>>>> The documentation doesn't say what happens if this limit is exceeded. >>>>> Also the default of 32758 suggests that we actually support that many >>>>> domains. I don't think we do... >>>>> >>>>> I didn't find anything in SUPPORT.md about how many guests we support >>>>> but I wouldn't want this setting here to imply full support for 32768 >>>>> domains. >>>>> >>>>> If you don't want to tackle this can of works, maybe add this: >>>>> >>>>> # This just controls some resource limits for xenstored; if the >>>>> # limit is exceeded, xenstored will stop being able to function >>>>> # properly for additional guests. The default value is so large >>>>> # that it won't be exceeded in a supported configuration, but >>>>> # should not be taken to mean that the whole Xen system is >>>>> # guaranteed to work properly with that many guests. >>>>> >>>>> Julien, did you ask for this to be made configurable ? Having written >>>>> the text above, I wonder if it wouldn't just be better to >>>>> unconditionally set it to "unlimited" rather than offering footgun >>>>> dressed up like a tuneable... >>>> >>>> So in v1 (see [1]), Juergen wanted to raise the limit. I assumed >>>> this meant that the default limit (configured by the system may not >>>> be enough). >>>> >>>> I felt this was wrong to impose an higher limit on everyone when an >>>> admin may know the maximum number of domains. >>>> >>>> By "unlimited", do you mean the calling "ulimit" (or whatever is >>>> used for configuring FDs) with unlimited? >>>> >>>> If so, I would be OK with that. My main was was to move the raising >>>> the limit outside Xenstored because: >>>> 1) This is easier for an admin to tweak it (in particular the OOM) >>>> 2) It feels wrong to me that the daemon chose the limits >>>> 3) An admin can enforce it >>> >>> Coming back to this series, I'm puzzled now. >>> >>> Julien, you didn't want me to raise the limit to a specific number >>> covering the maximum possible number of domains, because you thought >>> this might result in xenstored hogging huge numbers of file descriptors >>> in case of a bug. Why is unlimited better then? This will make the >>> possible number even larger. >> >> I don't think I suggested the unlimited number is better... My main >> objection in your original approach is you set an arbitrary limit you >> in Xenstored (which may not apply at all) and don't offer a way to the >> admin to tweak it. >> >> If the limit is set outside of Xenstored, then it becomes much easier >> for someone to just tweak the init script. I don't have a strong >> opinion on whether the default limit should be "unlimited" or a fixed >> number. > > xenstored is TCB. It needs a large number of FDs, and can be trusted > with unlimited. > > Also, like xenconsoled, we can calculate an upper bound, which is > derived from the ABI limit of 32k domids. > > All you're haggling over is the error semantics in the case of: > 1) the upper bound calculation is wrong, or > 2) there is an fd leak > > Personally, I think a fixed calculation is right, so fd leaks can be > spotted more obviously. > > An admin knob is not helpful - higher than the upper bound is just > wasteful, while lower will cause malfunctions. A lot of users will never run more than a few handful (maybe hundreds?) number of domains... So setting the limit to allow 32K of domains sound really wasteful and wouldn't really help to spot FD leaks. I don't really see the problem to allow an admin to say "I will never run more than N domains". So the limit can be lower and would still function normally. Do you mind explaining why you think otherwise? Cheers,
Andrew Cooper writes ("Re: [PATCH v3 2/2] tools/xenstore: set open file descriptor limit for xenstored"):
> xenstored is TCB. It needs a large number of FDs, and can be trusted
> with unlimited.
I baseically agree with this.
> Also, like xenconsoled, we can calculate an upper bound, which is
> derived from the ABI limit of 32k domids.
IMO the default should support at leaat this much.
However, I don't think you are right, because xenstored offers console
connections to (possibly arbitrarily many) local socket connections.
> All you're haggling over is the error semantics in the case of:
> 1) the upper bound calculation is wrong, or
> 2) there is an fd leak
>
> Personally, I think a fixed calculation is right, so fd leaks can be
> spotted more obviously.
>
> An admin knob is not helpful - higher than the upper bound is just
> wasteful, while lower will cause malfunctions.
I don't agree. Firstly, on a technical level, your statement is true
only if the admin does not know they will be running a much smaller
number of domains. Secondly, we have had several people saying they
want this to be configurable. I think if several people say they want
something to be configurable, we should respect that, even if we think
the use cases for it are marginal. If there are hazards in bad
settings of such a know, that can be dealt with in the docs.
Julien's point about not having the limit set by xenstored itself is
very well taken.
ISTM that the following scheme is in the intersection of everyone's
requirements:
* The limit will be adjusted/imposed in the startup script.
* An /etc/{default,sysconfig} parameter will be provided to
adjust the setting.
* The default should be `unlimtied` since we cannot calculate
a safe upper bound for all configurations.
* Systems like Citrix Hypervisor (XenServer) which can calculate
a safe upper bound can do so, and adjust the default, enabling
them to spot fd leaks.
Ian.
On 31.08.21 16:22, Ian Jackson wrote: > Andrew Cooper writes ("Re: [PATCH v3 2/2] tools/xenstore: set open file descriptor limit for xenstored"): >> xenstored is TCB. It needs a large number of FDs, and can be trusted >> with unlimited. > > I baseically agree with this. > >> Also, like xenconsoled, we can calculate an upper bound, which is >> derived from the ABI limit of 32k domids. > > IMO the default should support at leaat this much. > > However, I don't think you are right, because xenstored offers console > connections to (possibly arbitrarily many) local socket connections. > >> All you're haggling over is the error semantics in the case of: >> 1) the upper bound calculation is wrong, or >> 2) there is an fd leak >> >> Personally, I think a fixed calculation is right, so fd leaks can be >> spotted more obviously. >> >> An admin knob is not helpful - higher than the upper bound is just >> wasteful, while lower will cause malfunctions. > > I don't agree. Firstly, on a technical level, your statement is true > only if the admin does not know they will be running a much smaller > number of domains. Secondly, we have had several people saying they > want this to be configurable. I think if several people say they want > something to be configurable, we should respect that, even if we think > the use cases for it are marginal. If there are hazards in bad > settings of such a know, that can be dealt with in the docs. > > Julien's point about not having the limit set by xenstored itself is > very well taken. > > ISTM that the following scheme is in the intersection of everyone's > requirements: > > * The limit will be adjusted/imposed in the startup script. > * An /etc/{default,sysconfig} parameter will be provided to > adjust the setting. > * The default should be `unlimtied` since we cannot calculate > a safe upper bound for all configurations. > * Systems like Citrix Hypervisor (XenServer) which can calculate > a safe upper bound can do so, and adjust the default, enabling > them to spot fd leaks. Makes sense for me. So this would mean: - the sysconfig parameter will no longer be "number of domains", but the fd limit of the Xenstore daemon - default should be "unlimited" - the comment should mention the current number of fds needed per domain (5 for HVM, 2 for PV/PVH) plus some headroom in dom0 (without any guest running on my test system 21 fds are active, so I guess a value of 50 seems appropriate) Is this okay? Juergen
Juergen Gross writes ("Re: [PATCH v3 2/2] tools/xenstore: set open file descriptor limit for xenstored"):
> On 31.08.21 16:22, Ian Jackson wrote:
> > ISTM that the following scheme is in the intersection of everyone's
> > requirements:
> >
> > * The limit will be adjusted/imposed in the startup script.
> > * An /etc/{default,sysconfig} parameter will be provided to
> > adjust the setting.
> > * The default should be `unlimtied` since we cannot calculate
> > a safe upper bound for all configurations.
> > * Systems like Citrix Hypervisor (XenServer) which can calculate
> > a safe upper bound can do so, and adjust the default, enabling
> > them to spot fd leaks.
>
> Makes sense for me.
>
> So this would mean:
> - the sysconfig parameter will no longer be "number of domains", but the
> fd limit of the Xenstore daemon
> - default should be "unlimited"
> - the comment should mention the current number of fds needed per domain
> (5 for HVM, 2 for PV/PVH) plus some headroom in dom0 (without any
> guest running on my test system 21 fds are active, so I guess a value
> of 50 seems appropriate)
>
> Is this okay?
Sounds good to me, thanks.
Ian.
diff --git a/tools/hotplug/Linux/init.d/sysconfig.xencommons.in b/tools/hotplug/Linux/init.d/sysconfig.xencommons.in index 5ad4fe0818..2b682415f4 100644 --- a/tools/hotplug/Linux/init.d/sysconfig.xencommons.in +++ b/tools/hotplug/Linux/init.d/sysconfig.xencommons.in @@ -32,6 +32,13 @@ # Changing this requires a reboot to take effect. #XENSTORED=@XENSTORED@ +## Type: integer +## Default: 32768 +# +# Select maximum number of domains supported by xenstored. +# Only evaluated if XENSTORETYPE is "daemon". +#XENSTORED_MAX_N_DOMAINS=32768 + ## Type: string ## Default: "" # diff --git a/tools/hotplug/Linux/launch-xenstore.in b/tools/hotplug/Linux/launch-xenstore.in index 1747c96065..3f8b33dd32 100644 --- a/tools/hotplug/Linux/launch-xenstore.in +++ b/tools/hotplug/Linux/launch-xenstore.in @@ -54,6 +54,7 @@ test -f @CONFIG_DIR@/@CONFIG_LEAF_DIR@/xencommons && . @CONFIG_DIR@/@CONFIG_LEAF [ "$XENSTORETYPE" = "daemon" ] && { [ -z "$XENSTORED_TRACE" ] || XENSTORED_ARGS="$XENSTORED_ARGS -T @XEN_LOG_DIR@/xenstored-trace.log" + [ -z "$XENSTORED_MAX_N_DOMAINS" ] && XENSTORED_MAX_N_DOMAINS=32768 [ -z "$XENSTORED" ] && XENSTORED=@XENSTORED@ [ -x "$XENSTORED" ] || { echo "No xenstored found" @@ -63,6 +64,7 @@ test -f @CONFIG_DIR@/@CONFIG_LEAF_DIR@/xencommons && . @CONFIG_DIR@/@CONFIG_LEAF XS_OOM_SCORE=-$(($XENSTORED_OOM_MEM_THRESHOLD * 10)) rm -f @XEN_RUN_DIR@/xenstored.pid + N_FILES=$(($XENSTORED_MAX_N_DOMAINS * 5 + 100)) echo -n Starting $XENSTORED... $XENSTORED --pid-file @XEN_RUN_DIR@/xenstored.pid $XENSTORED_ARGS @@ -70,6 +72,7 @@ test -f @CONFIG_DIR@/@CONFIG_LEAF_DIR@/xencommons && . @CONFIG_DIR@/@CONFIG_LEAF systemd-notify --booted 2>/dev/null || timeout_xenstore $XENSTORED || exit 1 XS_PID=`cat @XEN_RUN_DIR@/xenstored.pid` echo $XS_OOM_SCORE >/proc/$XS_PID/oom_score_adj + prlimit --pid $XS_PID --nofile=$N_FILES exit 0 }
Add a configuration item for the maximum number of domains xenstored should support and set the limit of open file descriptors accordingly. For HVM domains there are up to 5 socket connections per domain (2 by the xl daemon process, and 3 by qemu). So set the ulimit for xenstored to 5 * XENSTORED_MAX_DOMAINS + 100 (the "+ 100" is for some headroom, like logging, event channel device, etc.). Signed-off-by: Juergen Gross <jgross@suse.com> --- V2: - set ulimit form launch script (Julien Grall) - split off from original patch (Julien Grall) --- tools/hotplug/Linux/init.d/sysconfig.xencommons.in | 7 +++++++ tools/hotplug/Linux/launch-xenstore.in | 3 +++ 2 files changed, 10 insertions(+)