Notice to HTCondor 7.8 users –
Statistics implemented during the 7.5 series that landed in 7.7.0 were rewritten by the time 7.8 was released. If you were using the original statistics for monitoring and/or reporting, here is a table to help you map old (left column) to new (right column).
See – 7.6 -> 7.8 schedd stats
(embedding content requires javascript, which is not available on wordpress.com)
Note: The *Rate and Mean* attributes require math, and UpdateTime requires memory
Configuration of accounting groups in HTCondor is too often an expert task that requires coordination between administrators and their tools.
Wallaby provides a coordination point, so long as a little convention is employed, and can provide a task specific interface to simplify configuration.
Quick background, Wallaby provides semantic configuration for HTCondor. It models a pool as parameters aggregated into features and nodes aggregated in groups, with features and individual parameters associated with nodes and groups. It provides semantic validation of configuration before it is distributed, and has expert knowledge for minimal impact configuration changes.
And, accounting group configuration in HTCondor is spread across seven fixed parameters (GROUP_NAMES, GROUP_ACCEPT_SURPLUS, GROUP_SORT_EXPR, GROUP_NAMES, GROUP_QUOTA_ROUND_ROBIN_RATE, GROUP_AUTOREGROUP, GROUP_QUOTA_MAX_ALLOCATION_ROUNDS), and another five dynamic parameters (GROUP_ACCEPT_SURPLUS_groupname, GROUP_AUTOREGROUP_groupname, GROUP_PRIO_FACTOR_groupname, GROUP_QUOTA_groupname, GROUP_QUOTA_DYNAMIC_groupname). These are dynamic because the “groupname” in the parameter is any name listed in the GROUP_NAMES parameter.
In addition to its other features, Wallaby has an extensible shell mechanism, which can be used to create task specific porcelain.
For instance, agree that tools and administrators will store accounting group configuration on a feature called AccountingGroups, and the tools can use Wallaby’s API to manipulate the configuration while the following porcelain can simplify the task for managing that configuration by administrators.
See – wallaby_accounting_group_porcelain.txt
(embedding content requires javascript, which is not available on wordpress.com)
A few years ago I discovered Web Numbr, a service that will monitor a web page for a number and graph that number over time.
I installed a handful of webnumbrs to track things at HTCondor’s gittrac instance.
http://webnumbr.com/search?query=condor
Thing such as –
I wanted to embed the numbers here, but javascript is needed and wordpress.com filters javascript from posts.
Concurrency limits allow for protecting resources by providing a way to cap the number of jobs requiring a specific resource that can run at one time.
For instance, limit licenses and filer access at four regional data centers.
CONCURRENCY_LIMIT_DEFAULT = 15 license.north_LIMIT = 30 license.south_LIMIT = 30 license.east_LIMIT = 30 license.west_LIMIT = 45 filer.north_LIMIT = 75 filer.south_LIMIT = 150 filer.east_LIMIT = 75 filer.west_LIMIT = 75
Notice the repetition.
In addition to the repetition, every license.* and filer.* must be known and recorded in configuration. The set may be small in this example, but imagine imposing a limit on each user or each submission. The set of users is board, dynamic and may differ by region. The set of submissions is a more extreme version of the users case, yet it is still realistic.
To simplify the configuration management for groups of limits, a new feature to provide group defaults to limit was added for the Condor 7.8 series.
The feature requires that only the exception to a rule be called out explicitly in configuration. For instance, license.west and filer.south are the exceptions in the configuration above. Simplified configuration available in 7.8,
CONCURRENCY_LIMIT_DEFAULT = 15 CONCURRENCY_LIMIT_DEFAULT_license = 30 CONCURRENCY_LIMIT_DEFAULT_filer = 75 license.west_LIMIT = 45 filer.south_LIMIT = 150
In action,
$ for limit in license.north license.south license.east license.west filer.north filer.south filer.east filer.west; do echo queue 1000 | condor_submit -a cmd=/bin/sleep -a args=1d -a concurrency_limits=$limit; done
$ condor_q -format '%s\n' ConcurrencyLimits -const 'JobStatus == 2' | sort | uniq -c | sort -n
30 license.east
30 license.north
30 license.south
45 license.west
75 filer.east
75 filer.north
75 filer.west
150 filer.south
If you don’t run tail -F on your logs periodically, you should. It’s illuminating. Try,
tail -F /var/log/condor/*Log | grep -i -e error -e fail -e warn
I ran that over the weekend and learned a few things –
0) ERROR WriteUserLog Failed to grab global event log lock means that the EVENT_LOG is lossy in unexpected ways. We know the EVENT_LOG rotates and if you’re watching it but miss a rotation you’ll miss events. However, when the above warning (not ERROR imho) is printed the event that was going to be written is dropped. So the EVENT_LOG could be lossy on the edges and in the middle.
1) GroupTracker (pid = 13252): fopen error: Failed to open file '/proc/13252/cgroup'. Error No such file or directory (2), coming from the ProcLog, means that a tracked process has disappeared. The exact implications are not clear, but the author, Brian Bockelman, suggest the message could be quieted as it doesn’t represent a functional problem. Maybe D_ALWAYS -> D_FULLDEBUG.
2) tail: `/var/log/condor/JobServerLog' has become inaccessible: No such file or directory many times in a row. When the job_queue.log is compressed, effectively recreated, the condor_job_server enters a phase where it reconstructs its internal state, in an apparently noisy fashion and can rotate its log file multiple times per second.
3) (1157197.152) (12639): attempt to connect to <10.10.10.10:52143> failed: Connection refused (connect errno = 111). and (1157197.152) (12639): Attempt to reconnect failed: Failed to connect to starter &tl;10.10.10.10:52143> turned out to be an issue on 10.10.10.10, where all jobs from a user were failing to start because of passwd_cache::cache_uid(): getpwnam("matt") failed: user not found with ERROR: Uid for "matt" not found in passwd file and SOFT_UID_DOMAIN is False and ERROR: Failed to determine what user to run this job as, aborting. The host was effectively a black hole because of a misconfigured UID_DOMAIN.
4) (1157079.244) (1199): ERROR "Can no longer talk to condor_starter <10.10.10.11:52725> turned out to be an issue on 10.10.10.11, where all jobs were failing to start because of Create_Process: Cannot access specified executable "/tmp/mycondor/release_dir/sbin/condor_starter": errno = 2 (No such file or directory) with slot5: ERROR: exec_starter failed! and slot5: ERROR: exec_starter returned 0, which was more bad configuration.
5) FileLock::obtain(1) failed - errno 0 (Success) looks wrong.
Physical machines are home to many types of resources these days. The traditional cores, memory, disk, now share space with gpus, co-processors or even protein sequence analysis accelerators.
To facilitate use and management of these resources, a new feature is available in HTCondor for extending machine resources. Analogous to concurrency limits, which operate on a pool / global level, machine resources operate on a machine / local level.
The feature allows a machine to advertise that it has specific types of resources available. Jobs can then specify that they require those specific types of resources. And the matchmaker will take into account the new resource types.
By example, a machine may have some GPU resources, an RS232 connected to your favorite telescope, and a number of physical spinning hard disk drives. The configuration for this would be,
MACHINE_RESOURCE_NAMES = GPU, RS232, SPINDLE MACHINE_RESOURCE_GPU = 2 MACHINE_RESOURCE_RS232 = 1 MACHINE_RESOURCE_SPINDLE = 4 SLOT_TYPE_1 = cpus=100%,auto SLOT_TYPE_1_PARTITIONABLE = TRUE NUM_SLOTS_TYPE_1 = 1
Aside – cpus=100%,auto instead of just auto because of GT3327. Also, the configuration for SLOT_TYPE_1 will likely go away in the future when all slots are partitionable by default.
Once a machine with this configuration is running,
$ condor_status -long | grep -i MachineResources MachineResources = "cpus memory disk swap gpu rs232 spindle" $ condor_status -long | grep -i -e TotalCpus -e TotalMemory -e TotalGpu -e TotalRs232 -e TotalSpindle TotalCpus = 24 TotalMemory = 49152 TotalGpu = 2 TotalRs232 = 1 TotalSpindle = 4 $ condor_status -long | grep -i -e ^Cpus -e ^Memory -e ^Gpu -e ^Rs232 -e ^Spindle Cpus = 24 Memory = 49152 Gpu = 2 Rs232 = 1 Spindle = 4
As you can see, the machine is reporting the different types of resources, how many of each it has and how many are currently available.
A job can take advantage of these new types of resources using a syntax already familiar for requesting resources from partitionable slots.
To consume one of the GPUs,
cmd = luxmark.sh request_gpu = 1 queue
Or for a disk intensive workload,
cmd = hadoop_datanode.sh request_spindle = 1 queue
With these jobs submitted and running,
$ condor_status
Name OpSys Arch State Activity LoadAv Mem ActvtyTime
slot1@eeyore LINUX X86_64 Unclaimed Idle 0.400 48896 0+00:00:28
slot1_1@eeyore LINUX X86_64 Claimed Busy 0.000 128 0+00:00:04
slot1_2@eeyore LINUX X86_64 Claimed Busy 0.000 128 0+00:00:04
Machines Owner Claimed Unclaimed Matched Preempting
X86_64/LINUX 3 0 2 1 0 0
Total 3 0 2 1 0 0
$ condor_status -l slot1@eeyore | grep -i -e ^Cpus -e ^Memory -e ^Gpu -e ^Rs232 -e ^Spindle
Cpus = 22
Memory = 48896
Gpu = 1
Rs232 = 1
Spindle = 3
That’s 22 cores, 1 gpu and 3 spindles still available.
Submit four more of the spindle consuming jobs and you’ll find the fourth does not run, because the available number of spindles is 0.
$ condor_status -l slot1@eeyore | grep -i -e ^Cpus -e ^Memory -e ^Gpu -e ^Rs232 -e ^Spindle Cpus = 19 Memory = 48512 Gpu = 1 Rs232 = 1 Spindle = 0
Since these custom resources are available as attributes in various ClassAds the same way Cpu, Memory and Disk are, all the policy, management and reporting capabilities you would expect is available.
The slot model was natural when a machine housed a single core. Though, the slot model did not exist when a machine housed a single core.
When machines were single core the model was a machine, represented as a MachineAd. A MachineAd had an associated CPU, some nominal amount of RAM and some chunk of disk space. Running a job meant consuming a machine.
When machines grew multiple cores the machine model was split. A single machine became independent MachineAds, called virtual machines. However, the name didn’t stick as the term virtual machine became a popular term in hardware virtualization. So a machine became independent MachineAds, called slots. The unifying entity, the machine itself, was lost. Running a job still meant consuming a slot.
Most recently, slots split into two classes: static and partitionable. Static slots are the slots formerly known as virtual machines. Partitionable slots are a representation of the physical machine itself, and are carved up, on-demand to service jobs. Both types are still MachineAds, but the consumption of partitionable slots is dynamic.
The slot model has demonstrated great utility but has been stretched.
In this time workloads have also changed. They have become more memory bound, disk IO bound, and network bound. They have started relying on specialized hardware and even application level services. They have started both spanning and packing into cores. They have grown complex data dependencies, become very short running, and become infrastructure level long running.
Machines have also grown to include scores of cores, hundreds of gigabytes of RAM, dozens of terabytes of disk, specialized hardware such as GPUs, co-processors, entropy keys, high speed interconnects and a bevy of other attached devices.
Machines are lumpy, heterogeneous means more than operating system and CPU architecture.
Furthermore, if it still existed, the machine model itself would fail to cleanly describe available resources. Classes of resources exist that house entire clusters, grids, or life-cycle manageable application services. Resources share addressable memory across operating systems instances, are custom architectures across whole data centers, and even those that don’t provide an outline of their capacity. Resources may grow and shrink while in use.
Consumption of these resources is not necessarily straightforward or uniform.
It’s time to stop thinking in slots. Its time to start thinking in aggregate resources and their consumption policies.
A question that often arises when approaching Condor from other batch systems is “How does Condor deal with resubmission of failed/preempted/killed jobs?”
The answer requires a slight shift in thinking.
Condor provides more functionality around the resubmission use case than most other schedulers. And the default policy is setup in such a way that most Condor folks don’t ever think about “resubmission.”
Condor will keep your job in the queue (condor_schedd managed) until the policy attached to the job says otherwise.
The default policy says a job will be run as many time as necessary for the job to terminate. So if the machine a job is running on crashes (generally, becomes unavailable), the condor_schedd will automatically try to run the job on another machine.
When you start changing the default policy you can control things such as: if a job should be removed after a period of time, even if it is running or only if it hasn’t started running; if a job should run multiple times even if it terminated cleanly; if a termination w/ an error should make the job run again, be held in the queue for inspection, be removed from the queue; if a job held for inspection should be held forever or a specific amount of time; if a job should only start running at a specific time in the future, or be run at repeated intervals.
The condor_submit manual page can provide specifics.
Condor has a few ways to run programs associated with a job, beyond the job itself. If you’re an administrator, you can use the USER_JOB_WRAPPER. If you’re a user who is friends with your administrator, you can use Job Hooks. If you are ambitious, you can wrap all your jobs in a script that runs programs before and after your actual job.
Or, you can use the PreCmd and PostCmd attributes on your job. They specify programs to run before and after your job executes. By example,
$ cat prepost.job cmd = /bin/sleep args = 1 log = prepost.log output = prepost.out error = prepost.err +PreCmd = "pre_script" +PostCmd = "post_script" transfer_input_files = pre_script, post_script should_transfer_files = always queue
$ cat pre_script #!/bin/sh date > prepost.pre $ cat post_script #!/bin/sh date > prepost.post
Running,
$ condor_submit prepost.job Submitting job(s) . 1 job(s) submitted to cluster 1. ...wait a few seconds, or 259... $ cat prepost.pre Sun Oct 14 18:06:00 UTC 2012 $ cat prepost.post Sun Oct 14 18:06:02 UTC 2012
That’s about it, except for some gotchas.
Condor produces internal data in both structured and unstructured forms.
The structured forms are just that and designed to be processed by external programs. These are the event logs (UserLog or EVENT_LOG), the HISTORY file and PER_JOB_HISTORY_DIR and POOL_HISTORY_DIR, and the job_queue.log and Accountantnew.log transaction logs.
The unstructured forms are for debugging and designed to be read by a person, often an experienced person. They are often called trace, or debug, logs and are the files in the LOG directory, or the extra output seen when passing -debug to command-line tools, i.e. condor_q -debug.
Consuming and processing the unstructured forms with external programs is increasingly important. Consider tracing incidents through a deployment of 50,000 geographically distributed, physical and virtual systems. Or, even 100 local systems.
More and more tools that provide the ability to aggregate unstructured logs are emerging and they all need to do some basic parsing of the logs. Help make their integration simpler and use a well defined format for timestamps.
For instance, ISO8601 –
DEBUG_TIME_FORMAT = "%Y-%m-%dT%H:%M:%S%z "
Spinning 