complex.5
NAME
complex - Grid Engine complexes configuration file format
DESCRIPTION
Complex reflects the format of the Grid Engine complex configuration.
The definition of complex attributes provides all pertinent information
concerning the resource attributes a user may request for a Grid Engine
job via the qsub(1) -l option, and for the interpretation of these
parameters within the Grid Engine system.
The Grid Engine complex object defines all entries which are used for
configuring the global, host, and queue objects. The system has a set
of pre-defined entries, which are assigned to a host or queue by
default. In addition, the user can define new entries and assign them
to one or more objects. Each load value has to have a corresponding
complex entry object, which defines the type and the relational
operator for it.
Defining resource attributes
The complex configuration should not be accessed directly. In order to
add or modify complex entries, the qconf(1) options -Mc and -mc should
be used instead. While the -Mc option takes a complex configuration
file as an argument and overrides the current configuration, the -mc
option brings up an editor filled in with the current complex
configuration.
The provided list contains all definitions of resource attributes in
the system. Adding a new entry means to provide: name, shortcut, type,
relop, requestable, consumable, default, and urgency. The fields are
described below. Changing one is easily done by updating the field to
change, and removing an entry by deleting its definition. An attribute
can only be removed when it is not referenced in a host or queue object
anymore. Also the system has a set of default resource attributes which
are always attached to a host or queue. They cannot be deleted, nor can
the type of such an attribute be changed.
Working with resource attributes
Before a user can request a resource attribute it has to be attached to
the global, host, or queue object. The resource attribute exists only
for the objects to which it was attached. If it is attached to the
global object (qconf -me global), it exists system-wide. Attached to a
host object (qconf -me host), it exists only on that host, and attached
to queue object (qconf -mq queue), only on that queue.
When an administrator attaches a resource attribute to an object, they
also have to assign a value to it: the resource limit. A load sensor
may be run to adjust the value presented by a host down from that
limit. For instance, to support requests for free space in the /tmp
filesystem, set up a load sensor to report the value (probably using
df(1)) and attach a sufficiently high limit to each host, e.g.
qconf -aattr exechost complex_values tmp_free=10T $(qconf -sel)
Default queue resource attributes
By default there is a selection of parameters in the queue
configuration as defined in queue_conf(5). The principal queue
configuration parameters requestable for a job by the user are:
qname
hostname
notify
calendar
min_cpu_interval
tmpdir
seq_no
s_rt
h_rt
s_cpu
h_cpu
s_data
h_data
s_stack
h_stack
s_core
h_core
s_rss
h_rss
Default host resource attributes
The standard set of host-related attributes consists of two categories.
The first category is built by several queue configuration attributes
which are particularly suitable to be managed on a host basis. These
attributes are:
slots
s_vmem
h_vmem
s_fsize
h_fsize
(Please refer to queue_conf(5) for details.)
Note: Defining these attributes in the host complex is no contradiction
to having them also in the queue configuration. It allows maintaining
the corresponding resources on a host level, and at the same time on a
queue level. Total virtual free memory (h_vmem) can be managed for a
host, for example, and a subset of the total amount can be associated
with a queue on that host.
The second attribute category in the standard host complex is that of
the default load values every sge_execd(8) periodically reports load to
sge_qmaster(8). The reported load values are either the standard Grid
Engine load values, such as the CPU load average (see uptime(1)), or
load values defined by the Grid Engine administration (see the
load_sensor parameter in the cluster or host configuration (see
sge_conf(5) for details). The definition of characteristics for the
standard load values is part of the default host complex, while
administrator-defined load values require extension of the host
complex. Please refer to load_parameters(5) for detailed information on
the standard set of load values.
Overriding attributes
An attribute can be assigned to the global object, host object, and
queue object at the same time. On the host level it might get its value
from the user-defined resource limit and a load sensor. If the
attribute is a consumable, we have, in addition to the resource limit
and its load report at host level, also the internal usage which the
system keeps track of. The merge is done as follows:
In general an attribute can be overridden on a lower level
- global by hosts and queues
- hosts by queues and load values or resource limits on the same
level.
We have one limitation for overriding attributes based on their
relational operator:
!= and == operators can only be overridden on the same level, not on a
lower level. The user-defined value always overrides the load value.
>=, >, <=, and < operators can only be overridden when the new value is
more restrictive than the old one.
In the case of a consumable at host level which has also a load sensor,
the system checks for the current usage, and if the internal accounting
is more restrictive than the load sensor report, the internal value is
kept; if the load sensor report is more restrictive, that one is kept.
FORMAT
The principal format of a complex configuration is that of a tabulated
list. Each line starting with a '#' character is a comment line. Each
non-comment line defines one element of the complex. Backslashes (\)
be used to escape newline characters. The backslash and the newline are
replaced with a space character before any interpretation.
An element definition line consists of the following 8 column entries
per line (in order of appearance):
name
The name of the complex element to be used to request this attribute
for a job in the qsub(1) -l option. A complex attribute name (see
complex_name in sge_types(5)) may appear only once across all
complexes, i.e. the complex attribute definition is unique.
shortcut
A shortcut for name which may also be used to request this attribute
for a job in the qsub(1) -l option. A given shortcut may appear only
once across all complexes, so as to avoid the possibility of ambiguous
complex attribute references.
type
This setting determines how the corresponding values are to be treated
by Grid Engine internally in comparisons or in load scaling for the
load complex entries:
o With INT only raw integers are allowed.
o With DOUBLE floating point numbers in double precision (decimal and
scientific notation) can be specified.
o With TIME time specifiers are allowed. Refer to sge_types(5) for a
format description.
o With MEMORY memory size specifiers are allowed. Refer to
sge_types(5) for a format description.
o With BOOL the strings TRUE and FALSE are allowed. When used in a
load formula (refer to sched_conf(5)), TRUE and FALSE get mapped
into '1' and '0'.
o With STRING all strings are allowed and are used for wildcard
regular boolean expression matching. Please see the sge_types(5)
man page for expression definition.
Examples:
-l arch="*x*|sol*" :
results in "arch=lx-x86" OR "arch=lx-amd64"
OR "arch=sol-amd64" OR ...
-l arch="sol-x??" :
results in "arch=sol-x86" OR "arch=sol-x64" OR ...
-l arch="lx2[246]-x86" :
results in "arch=lx22-x86" OR "arch=lx24-x86"
OR "arch=lx26-x86"
-l arch="lx2[4-6]-x86" :
results in "arch=lx24-x86" OR "arch=lx25-x86"
OR "arch=lx26-x86"
-l arch="lx2[24-6]-x86" :
results in "arch=lx22-x86" OR "arch=lx24-x86"
OR "arch=lx25-x86" OR "arch=lx26-x86"
-l arch="!lx-x86&!sol-amd64" :
results in NEITHER "arch=lx-x86" NOR "arch=sol-amd64"
-l arch="lx2[4|6]-amd64" :
results in "arch=lx24-amd64" OR "arch=lx26-amd64"
o CSTRING is like STRING except comparisons are case insensitive.
o RESTRING is the same as STRING for historical compatibility, but is
deprecated and may be removed in future..
o HOST is like CSTRING but the expression must match a valid host
name.
relop
The relation operator is used when the value requested by the user for
this parameter is compared against the corresponding value configured
for the considered queues. If the result of the comparison is false,
the job cannot run in this queue. Possible relation operators are "==",
"<", ">", "<=", ">=" and "EXCL". The only valid operator for string
type attributes is "==".
The "EXCL" relation operator implements exclusive scheduling and is
only valid for consumable boolean type attributes. Exclusive means the
result of the comparison is only true if a job requests to be
exclusive, and no other exclusive or non-exclusive job uses the
complex. If the job does not request to be exclusive and no other
exclusive job uses the complex the comparison is also true.
requestable
The entry can be used in a qsub(1) resource request if this field is
set to 'y' or 'yes'. If set to 'n' or 'no' this entry cannot be used
by a user in order to request a queue or a class of queues. If the
entry is set to 'forced' or 'f' the attribute has to be requested by a
job, or it is rejected.
To enable resource request enforcement the existence of the resource
has to be defined. This can be done on a cluster global, per host and
per queue basis. The definition of resource availability is performed
with the complex_values entry in host_conf(5) and queue_conf(5).
consumable
The consumable parameter can be set to either 'yes' ('y' abbreviated),
'no' ('n') or 'JOB' ('j'). It can be set to 'yes' and 'JOB' only for
numeric attributes (INT, DOUBLE, MEMORY, TIME - see type above). If set
to 'yes' or 'JOB' the consumption of the corresponding resource can be
managed by Grid Engine internal bookkeeping. In this case Grid Engine
accounts for the consumption of this resource for all running jobs and
ensures that jobs are only dispatched if the Grid Engine internal
bookkeeping indicates enough available consumable resources.
Consumables are an efficient means to manage limited resources such as
available memory, free space on a file system, network bandwidth or
floating software licenses.
A consumable defined by 'y' is a per-slot consumable, which means the
limit is multiplied by the number of slots being used by the job before
being applied. In case of 'j' the consumable is a per-job consumable.
This resource is debited as requested (without multiplication) from the
allocated master queue. The resource need not be available for the
slave task queues.
Consumables can be combined with default or user-defined load
parameters (see sge_conf(5) and host_conf(5)), i.e. load values can be
reported for consumable attributes, or the consumable flag can be set
for load attributes. The Grid Engine consumable resource management
takes both the load (measuring availability of the resource) and the
internal bookkeeping into account in this case, and makes sure that
neither exceeds a given limit.
To enable consumable resource management, the basic availability of a
resource has to be defined. This can be done on a cluster global, per
host and per queue basis, and these categories may supersede each other
in the given order (i.e. a host can restrict availability of a cluster
resource and a queue can restrict host and cluster resources). The
definition of resource availability is performed with the
complex_values entry in host_conf(5) and queue_conf(5). The
complex_values definition of the "global" host specifies cluster global
consumable settings. To each consumable complex attribute in a
complex_values list, a value is assigned which denotes the maximum
available amount for that resource. The internal bookkeeping will
subtract from this total the assumed resource consumption by all
running jobs as expressed through the jobs' resource requests.
Note: Jobs can be forced to request a resource and thus to specify
their assumed consumption via a forced value of the requestable
parameter (see above).
Note also: A default resource consumption value can be pre-defined by
the administrator for consumable attributes not explicitly requested by
the job (see the default parameter below). This is meaningful only if
requesting the attribute is not enforced as explained above.
default
Meaningful only for consumable complex attributes (see consumable
parameter above) and must be specified as 0 otherwise. Grid Engine
assumes the resource amount denoted in the default parameter implicitly
to be consumed by jobs being dispatched to a host or queue managing the
consumable attribute. Jobs explicitly requesting the attribute via the
-l option to qsub(1) override this default value.
urgency
The urgency value allows influencing job priorities on a per-resource
base. The urgency value effects the addend for each resource when
determining the resource request-related urgency contribution. For
numeric type resource requests the addend is the product of the urgency
value, the job's assumed slot allocation, and the per-slot request as
specified via the -l option to qsub(1). For string type requests the
resource's urgency value is directly used as addend. Urgency values are
of type real. See under sge_priority(5) for an overview of job
priorities.
SEE ALSO
sge_intro(1), sge_types(1), qconf(1), qsub(1), uptime(1), host_conf(5),
load_parameters(5), queue_conf(5), sge_execd(8), sge_qmaster(8)
COPYRIGHT
See sge_intro(1) for a full statement of rights and permissions.
SGE 8.1.3pre 2011-12-04 COMPLEX(5)
Man(1) output converted with
man2html