2,857 questions
Advice
0
votes
2
replies
57
views
How to make devices discover each other using WIFI
What's the best way to allow programs to discover each other on the network?
Let's say we are writing a system that tracks the usage of computers over the network.
We have an agent program that sends ...
- 13
Advice
2
votes
2
replies
59
views
Efficient MPI Parallelization Strategies for Localized PDE Subproblems within a Globally Decomposed Domain
I am working on a global PDE problem that is solved using a standard domain-decomposition strategy (e.g., Scotch, METIS). This part of the computation is well balanced across all MPI processes.
...
- 1
0
votes
1
answer
44
views
Upsert! Operation Throws "A table can't contain duplicate column names" Error
I have a base table A and a result table B in DolphinDB. Table B was initially empty and is used to store calculated results based on table A. When trying to insert the calculated results into table B,...
- 1
0
votes
0
answers
157
views
vLLM + Ray multi-node tensor-parallel deployment completely blocked by pending placement groups and raylet heartbeat failures
Environment:
Ray version: 2.x
vLLM version: 0.9.2
Python version: 3.9
OS / Container base: Linux (CentOS-based UBI8 in Kubernetes)
Cloud / Infrastructure: AWS based Kubernetes cluster (pods scheduled ...
- 9
3
votes
1
answer
139
views
In Apache Ignite the Replication mode and Partition mode does not work all together
I’m working with Apache Ignite 2.17.0. I load database tables into Ignite caches and run SQL queries using the SQLFieldsQuery API.
Recently, I modified the cache configuration for some tables to use ...
- 799
0
votes
0
answers
64
views
Get two different nodes to access and distribute the same SQL table in Apache spark?
I have the following code to test. I created a table on worker 1. Then I tried to read the table on worker 2 and it got TABLE_OR_VIEW_NOT_FOUND. Worker 2 is in the some computer as Master.
I ran the ...
3
votes
2
answers
284
views
How Ray async actors handle calls to sync methods
I'm working with Ray async actors and I want to understand exactly what happens—at a deep technical level—when a synchronous method is called on such an actor.
I know that calling a synchronous method ...
- 893
0
votes
0
answers
52
views
How to best partition my data with a 32 core EMR instance and make sure I max out the parallelize feature?
I’m optimizing a PySpark pipeline that processes records with a heavily skewed categorical column (category). The data has:
A few high-frequency categories (e.g., 90% of records fall into 2-3 ...
- 27
0
votes
0
answers
154
views
How to set up MS-MPI multi-machine communication between two Windows 11 systems?
I'm trying to set up a multi-machine communication environment using MS-MPI on two Windows 11 laptops, but I'm encountering some issues. Here are the details of my setup:
Environment Details:
...
1
vote
1
answer
135
views
Distributed REST API Calls using SPARK with maintaining consistency
I have a Spark DataFrame created from a Delta table, with one column of type STRUCT(JSON). For each row in this DataFrame, I need to make a REST API call using the JSON payload in the column. ...
- 53
0
votes
0
answers
21
views
MLP Speed-Up in PySpark fluctuates with more cores – possible cache memory issue?
enter image description here
I have conducted experiments running the MLP (Multi-Layer Perceptron) algorithm on a PC cluster with Apache Spark, with configurations ranging from small data to large ...
0
votes
0
answers
336
views
PyTorch DDP Multi-Node Training: ncclInternalError: Internal check failed. Bootstrap : no socket interface found
I am trying to run a multi-node training job using PyTorch's DistributedDataParallel (DDP) following this guide. However, when I launch the job with torchrun, I encounter the following NCCL error on ...
- 1
0
votes
1
answer
983
views
Clearing Cached Data on Databricks Cluster
The problem I am facing is that my "used" memory is only around 16GB, however the cached memory takes up so much space, that I am forced to use a compute with higher memory (64GB).
So I ...
1
vote
0
answers
96
views
Segmentation Fault During Validation with MirroredStrategy on Multiple GPUs
I am training a model using TensorFlow 2.18.0 with the tf.distribute.MirroredStrategy across two GPUs. The training works fine on a single GPU, but when I try to run it on two GPUs, it ends with a ...
- 56
0
votes
1
answer
117
views
I want to use the distributed package in PyTorch for point-to-point communication between two ranks. but run error
def runTpoly(rank, size, pp, cs, pkArithmetics_evals,
pkSelectors_evals, domain):
init_process(rank, size)
group2 = torch.distributed.new_group([1,2])
if rank == 0:
device ...
0
votes
0
answers
82
views
Vertex AI Reduction Server returning 500 Internal Error
I am looking to finetune a pre-trained deberta model on Vertex AI with pytorch. I'm attempting to run a distributed job, making use of the Vertex AI reduction server.
I'm following this notebook: ...
0
votes
0
answers
66
views
How to develop multi-GPU modules in single-node single-GPU system in pytorch?
I'm developing a multi-GPU PyTorch application. Existing methods like scatter/gather in torch.distributed don't fulfill my requirements, therefore I need to develop forward/backprop steps which send ...
1
vote
0
answers
36
views
Distributed training with Trainer and ConstantLengthDataset classes
I have a custom ConstantLengthDataset class:
class ConstantLengthDataset(IterableDataset):
def __init__(
self,
tokenizer,
dataset,
infinite=False,
...
- 11
0
votes
2
answers
65
views
Interrupting busy worker process
I am trying to benchmark an algorithm I developed. To this end, I run the algorithm on several instances and measure time, memory, other numbers... For each instance, I create a new process; partly ...
- 893
0
votes
1
answer
41
views
How to maintain synchronization between distributed python processes?
I have a number of workstations that run long processes containing sequences like this:
x = wait_while_current_is_set
y = read_voltage
z = z + y
The workstations must maintain synchronization with a ...
- 2,706
0
votes
1
answer
498
views
Writing to a delta table spark 3.5.3 delta lake 3.2.0
It seems I'm unable to write using the delta format from my spark job, but I'm not sure what I'm missing. I'm using spark 3.5.3 and deltalake 3.2.0.
My error:
Exception in thread "main" org....
- 141
0
votes
1
answer
232
views
How to configure Ray cluster to utilize the Full Capacity of Databricks Cluster
I have a Databricks cluster configured with a minimum of 1 worker and a maximum of 4 workers, with auto-scaling enabled. What should my Ray configuration (setup_ray_cluster) be to fully utilize the ...
- 3,018
0
votes
0
answers
383
views
Cuda failure ‘named symbol not found’ when run on 4 L4 GPUs
Can someone help me with the following error. The code works fine on the 2 T4 GPUs. But fails when run on the 4 L4 GPUs. I am extending the Gemma 2B model for a multi-label multi-class classification ...
1
vote
0
answers
92
views
function to count maximum number of co-occurring entities for combination of 3.5 million ids?
I have been given this:
x | y
1 | a,b,c,d,e
2 | a,b,c,d
3 | a,c,d
...
I'd like this:
1,2 | 4 (a,b,c,d)
1,3 | 3 (a,c,d)
2,3 | 3 (a,c,d)
I have x -> 3*10^6 such rows (3 million records)
y could ...
1
vote
0
answers
22
views
Why am I getting "TXN_REQUEST_IGNORED ERROR 10906" in GridDB due to an unknown event during cluster operations?
I’m using GridDB for a distributed database setup and recently encountered the following error while performing operations across nodes in the cluster:
from griddb_python import StoreFactory, ...
0
votes
1
answer
73
views
Fault-tolerant queue-worker architecture in Kafka?
I am new to using queue-worker architectures and I'm interested in how to make it resilient to a worker failing. For example
We have a pool of workers Alpha that put entries onto queue A
Then the ...
- 178
2
votes
2
answers
226
views
Client request failure in raft
Imagine a 3 node raft cluster. Each node is in sync has log [1,2,3] and entry 3 is committed by the leader.
Now leader receives an entry 4 but fails to commit it because of unreliable network and ...
- 129
2
votes
0
answers
36
views
Why am I getting "SYNC_LOG_NOT_FOUND ERROR 20021" in GridDB during cluster synchronization?
I'm working with GridDB to manage a distributed cluster, and recently I’ve been encountering the following error during data synchronization across nodes in the cluster:
20021 SYNC_LOG_NOT_FOUND ERROR
...
0
votes
1
answer
144
views
Apache ignite compute broadcast
I'm trying apache ignite and must say ignite documentation is incomplete. Anyway, I've setup two node cluster using docker images 2.14.0-arm14 and exposed all Ports for both ignite containers, however ...
- 196
1
vote
0
answers
81
views
Dask erring on GridSearchCV and RandomizedSearchCV
I am trying to train an xgboost model using dask. I already have the data transformed andhave prepared my data as follows:
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, ...
- 65
2
votes
0
answers
432
views
How to Implement Distributed AutoFAISS with PySpark for Large-Scale Vector Indexing?
I’m working on a project that involves creating a vector search index for a massive dataset consisting of 1.3 trillion tokens. I want to use AutoFAISS in a distributed environment to handle the scale ...
- 2,759
1
vote
1
answer
57
views
run a function on different nodes of a slurm cluster for different parameters
How do I call the function inner from outer, such that each call to inner runs on a different node? That is, for ij = 1, it runs on node 1 using all of its 16 cores, for ij = 2, it runs on node 2 ...
0
votes
1
answer
113
views
Do I need Apache Spark to execute my Airflow DAG tasks?
I have a workflow with multiple DAGs. Every DAG has multiple tasks.
These tasks are simple ETL tasks. It involves geo data in the form of kmls, csvs.
An example task:
We have meta data of road ...
0
votes
1
answer
56
views
reliable protocol guarantee complete delivery no in order promise
The sender is sending N packets to receiver.
I want a protocol or method that guarantees delivery, each packet is received at least once. It is ok if some packets are received more than once due to ...
- 15
0
votes
0
answers
35
views
How to securely conduct lottery-like draws with guaranteed randomness without auditing?
Is there an existing algorithm or method to conduct lottery-like draws that ensures secure and truly random results without the need for auditing?
There are any lib to do this?
I search on the web ...
- 233
0
votes
2
answers
552
views
Unable to run code on Multiple GPUs in PyTorch - Usage shows only 1 GPU is being utilized
I am training a Transformer Encoder-Decoder based model for Text summarization. The code works without any errors but uses only 1 GPU when checked with nvidia-smi. However, I want to run it on all the ...
- 11
1
vote
2
answers
958
views
How to reliably implement fan out write pattern?
I'm trying to RELIABLY implement that pattern.
For practical purposes, assume we have something similar to a twitter clone (in cassandra and nodejs).
So, user A has 500k followers. When user A posts a ...
1
vote
1
answer
418
views
Using torchrun with AWS sagemaker estimator on multi-GPU node
I would like to run a training job ml.p4d.24xlarge machine on AWS SageMaker. I ran into a similar issue described here with significant slowdowns in training time. I understand now that I should run ...
- 23
2
votes
0
answers
26
views
Why am I getting a "LM_WRITE_LOG_FAILED ERROR 80000" in GridDB when writing to the log file?
I'm using GridDB for managing a distributed database system and recently encountered the following error while trying to perform operations:
80000 LM_WRITE_LOG_FAILED ERROR
Writing to log file failed. ...
- 91
2
votes
0
answers
25
views
Why am I getting a "JC_CONTAINER_NOT_OPENED ERROR 145034" in GridDB when performing operations on a container?
I'm working with GridDB to manage a distributed database and recently encountered the following error while performing operations on a container:
145034 JC_CONTAINER_NOT_OPENED ERROR Status check of ...
- 91
2
votes
0
answers
45
views
Why am I getting a "SYNC_CREATE_CONTEXT_FAILED ERROR 20037" during data synchronization in my GridDB cluster?
I'm working on a distributed system where I need to synchronize data across a cluster of nodes. However, I'm encountering an error during the synchronization process. The error message I get is:
20037 ...
- 91
3
votes
1
answer
77
views
Async leader election in unrooted spanning tree declares multiple winners
I am trying to implement the algorithm described in the image, using MPI. It is part of a University project where we are building a distributed satellite to ground station communication system. I ...
- 1,353
1
vote
0
answers
106
views
How to properly clean up non-serializable states associated with a Ray object?
Suppose I have a Ray actor that can create a Ray object that associates with some non-serializable states. In the following example, the non-serializable state is a temporary directory.
class MyObject:...
- 3,773
1
vote
1
answer
124
views
Error: invalid child spec in supervisor start_child function
In my erlang application i have a top level supervisor that monitors a cowboy server (gen_server):
start_link() ->
supervisor:start_link({local, ?SERVER}, ?MODULE, []).
init([]) ->
...
- 71
2
votes
0
answers
104
views
How does XGBoost aggregate models being trained in a distributed fashion across n machines?
I am trying to understand how XGBoost distributed training works. The best explanation I've found so far is in this paper: https://ml-pai-learn.oss-cn-beijing.aliyuncs.com/%E6%9C%BA%E5%99%A8%E5%AD%A6%...
- 359
0
votes
2
answers
118
views
Why does decreasing partition count prevent a StageFailure due to large size of serialized results?
I am trying to generate a randomly sorted version of a large-ish dataframe on databricks.
My go-to code is to use .orderBy(rand()) on the dataframe.
This, however, seems to trigger a SparkException ...
- 11.9k
0
votes
1
answer
168
views
Micrometer & Prometheus with Java subprocesses that can't expose HTTP
I have a complex product that runs like this.
A parent Java process which expose an HTTP service.
The parent process starts worker subprocesses (new JVM) and manage the lifecycle of them.
Worker ...
- 510
0
votes
1
answer
110
views
Least Connection Load balancing using Grpc
least_connection.proto code
Node overloaded -- starting load balancing process
Traceback (most recent call last):
File "D:\lab7p2\least connection\node2.py", line 73, in <module>
node....
1
vote
1
answer
2k
views
How to debug ValueError: `FlatParameter` requires uniform dtype but got torch.float32 and torch.bfloat16?
I'm trying to do Pytorch Lightning Fabric distributed FSDP training with Huggingface PEFT LORA fine tuning on LLAMA 2 but my code ends up failing with:
`FlatParameter` requires uniform dtype but got ...
- 2,376
0
votes
1
answer
53
views
Load pre-training parameters trained on a single GPU on multi GPUS on a single machine
I tried to load the pre-training parameters trained by a single GPU on a single machine with multiple GPUs, but errors such as Missing keys and Unexpected keys occurred.
backbone_cfg = dict(
...
