I am storing logs into elastic search from my reactive spring application. I am getting the following error in elastic search:
Elasticsearch exception [type=es_rejected_execution_exception, reason=rejected execution of processing of [129010665][indices:data/write/bulk[s][p]]: request: BulkShardRequest [[logs-dev-2020.11.05][1]] containing [index {[logs-dev-2020.11.05][_doc][0d1478f0-6367-4228-9553-7d16d2993bc2], source[n/a, actual length: [4.1kb], max length: 2kb]}] and a refresh, target allocation id: WwkZtUbPSAapC3C-Jg2z2g, primary term: 1 on EsThreadPoolExecutor[name = 10-110-23-125-common-elasticsearch-apps-dev-v1/write, queue capacity = 200, org.elasticsearch.common.util.concurrent.EsThreadPoolExecutor#6599247a[Running, pool size = 2, active threads = 2, queued tasks = 221, completed tasks = 689547]]]
My index settings:
{
"logs-dev-2020.11.05": {
"settings": {
"index": {
"highlight": {
"max_analyzed_offset": "5000000"
},
"number_of_shards": "3",
"provided_name": "logs-dev-2020.11.05",
"creation_date": "1604558592095",
"number_of_replicas": "2",
"uuid": "wjIOSfZOSLyBFTt1cT-whQ",
"version": {
"created": "7020199"
}
}
}
}
}
I have gone through this site:
https://www.elastic.co/blog/why-am-i-seeing-bulk-rejections-in-my-elasticsearch-cluster
I thought adjusting "write" size in thread-pool will resolve, but it is mentioned as not recommended in the site as below:
Adjusting the queue sizes is therefore strongly discouraged, as it is like putting a temporary band-aid on the problem rather than actually fixing the underlying issue.
So what else can we do improve the situation?
Other info:
Elastic Search version 7.2.1
Cluster health is good and they are 3 nodes in cluster
Index will be created on daily basis, there are 3 shards per index
While you are right, that increasing the thread_pool size is not a permanent solution, you will be glad to know that elasticsearch itself increased the size of write thread_pool(use in your bulk requests) from 200 to 10k in just a minor version upgrade. Please see the size of 200 in ES 7.8, while 10k of ES 7.9 .
If you are using the ES 7.X version, then you can also increase the size to if not 10k, then at least 1k(to avoid rejecting the requests).
If you want a proper fix, you need to do the below things
Find out if it's consistent or just some short-duration burst of write requests, while gets cleared in some time.
If it's consistent, then you need to figure out if have all the write optimization is in place, please refer to my short-tips to improve index speed.
See, if you have reached the full-capacity of your data-nodes, and if yes, scale your cluster to handle the increased/legitimate load.
Related
We recently upgraded our elastic cloud deployment from 6.8.5 to 7.9
After the upgrade we are seeing the following error time to time.
{
"error" : {
"root_cause" : [
{
"type" : "circuit_breaking_exception",
"reason" : "[parent] Data too large, data for [<http_request>] would be [416906520/397.5mb], which is larger than the limit of [408420352/389.5mb], real usage: [416906520/397.5mb], new bytes reserved: [0/0b], usages [request=0/0b, fielddata=32399/31.6kb, in_flight_requests=0/0b, model_inference=0/0b, accounting=4714192/4.4mb]",
"bytes_wanted" : 416906520,
"bytes_limit" : 408420352,
"durability" : "PERMANENT"
}
],
"type" : "circuit_breaking_exception",
"reason" : "[parent] Data too large, data for [<http_request>] would be [416906520/397.5mb], which is larger than the limit of [408420352/389.5mb], real usage: [416906520/397.5mb], new bytes reserved: [0/0b], usages [request=0/0b, fielddata=32399/31.6kb, in_flight_requests=0/0b, model_inference=0/0b, accounting=4714192/4.4mb]",
"bytes_wanted" : 416906520,
"bytes_limit" : 408420352,
"durability" : "PERMANENT"
},
"status" : 429
}
This deployment consists of only one node with 1G memory. We would like to know the cause of this error. Is it due to the upgrade?
Thank you.
First, the circuit breaker is a protection that some request doesn't push your cluster over the limit of what it can handle — this is killing a single request rather than (potentially) the entire cluster. Also note that this HTTP request alone isn't too large, but it trips the parent circuit breaker — so this request on top of everything else would be too much.
The initial circuit breaker was already added in 6.2.0, but was tightened down further in 7.0.0. I assume that's the reason why you are seeing this (more frequently) now.
You could change the indices.breaker.total.limit, but this isn't a magic switch to get more out of your cluster. 1GB of memory might just not be enough for what you are trying to do.
I have multiple path.datas configured for my Elasticsearch cluster.
The official documentation states that only a single path is used for a single shard, so it's never splitted across multiple paths.
I'd like to find a way to finding out which path on which node is used for some specific shard (primary or replica), like index my-index primary shard 0 → node RQzJvAgLTDOnEnmIjYU9FA path /mnt/data1. Tried /_nodes, /_stats, /_segments, /_shard_stores, but there are no any references to paths.
You can find that info using the indices stats API by specifying the level=shards parameter
GET index/_stats?level=shards
will return a structure like this
"indices": {
"listings-master": {
"primaries": {
...
},
"total": {
...
},
"shards": {
"0": [
{
"shard_path": {
"state_path": "/app/data/nodes/0",
"data_path": "/app/data/nodes/0",
"is_custom_data_path": false
},
...
}
...
Not easily but but by doing a small python script I've the info I want, here the script
import json
with open('shard.json') as json_file:
data = json.load(json_file)
print(data.keys())
data=data['indices']
for indice in data:
#print(indice)
d1=data[indice]
shards=d1['shards']
#print(shards,type(shards),shards.keys())
for nshard in shards.keys():
shard=shards[nshard]
#print(shard,type(shard))
for elt in shard:
path=elt['shard_path']['data_path']
node=elt['routing']['node']
#print(repr(elt['shard_path']['data_path']))
#print("=========================")
print(indice,'\t',nshard,'\t',node,'\t',path)
They you obtain stuff like
log-2020.11.06 1 oxx /datassd/elasticsearch/nodes/0
log-2020.11.06 0 oxx /datassd/elasticsearch/nodes/0
log-2020.11.05 1 oxx /datassd/elasticsearch/nodes/0
I have a ops cluster with 5 nodes (1 master, 1 client, and 3 data nodes). I want to restore a backup of this onto a new test cluster with only 3 nodes (1 master, 1 client, 1 data). I only have 1 data node in my test cluster at the moment and wasn't planning to add any additional data nodes on my test cluster.
The issue I'm having is that when I try to restore to my test cluster, only some of the shards get assigned. Most of them stay in the UNASSIGNED state. I've tried to use the reroute api but it fails. See below
Does my test cluster have to have the same number of nodes as my ops cluster I'm restoring from? If so is there any work around for this?
{
"error": {
"root_cause": [
{
"type": "reroute_transport_exception",
"reason": ["myhost_master"[myhostip:9200][cluster:admin/reroute]"
}
],
"type": "illegal_argument_exception",
"reason": "resovled [myhostip] into [3] nodes, where excpeted to be resolved to a single node"
],
"status": 400
}
I have got a index with 5 primary shards and no replicas.
One of my shard(shard 1) is in unassigned state. When i checked the log file, i found out below error:
2obv65.nvd, _2vfjgt.fdx, _3e3109.si, _3dwgm5_Lucene45_0.dvm, _3aks2g_Lucene45_0.dvd, _3d9u9f_76.del, _3e30gm.cfs, _3cvkyl_es090_0.tim, _3e309p.nvd, _3cvkyl_es090_0.blm]]; nested: FileNotFoundException[_101a65.si]; ]]
When i checked the index, i could not find the 101a65.si file for the shard 1.
I am unable to locate the missing .si file. I tried a lot but could not assign the shard 1 again.
Is there any other way to make the shard 1 assign again? or do i need to delete the entire shard 1 data?
Please suggest.
Normally in the stack trace you should see the path to the corrupted shard, something like MMapIndexInput(path="path/to/es/db/nodes/node_number/indices/name_of_index/1/index/some_file) (here the 1 is the shard number)
Normally deleting path/to/es/db/nodes/node_number/indices/name_of_index/1 should help the shard recover. If you still see it unassigned try sending this command to your cluster (normally as per the documentation, it should work, though I'm not sure about ES 1.x syntax and commands):
POST _cluster/reroute
{
"commands" : [
{
"allocate" : {
"index" : "myIndexName",
"shard" : 1,
"node" : "myNodeName",
"allow_primary": true
}
}
]
}
I want to index documents into Elasticsearch from Storm, but I couldn't get any document to be indexed into Elasticsearch.
In my topology I have a KafkaSpout that emits a json like this { “tweetId”: 1, “text”: “hello” } to a EsBolt that is a native bolt from elasticsearch-hadoop library that writes the Storm Tuples to Elasticsearch (doc is here: https://www.elastic.co/guide/en/elasticsearch/hadoop/current/storm.html).
These are the configs for my EsBolt:
Map conf = new HashMap();
conf.put("es.nodes","127.0.0.1");
conf.put("es.port","9200");
conf.put("es.resource","twitter/tweet");
conf.put("es.index.auto.create","no");
conf.put("es.input.json", "true");
conf.put("es.mapping.id", "tweetId");
EsBolt elasticsearchBolt = new EsBolt("twitter/tweet", conf);
The first two configurations have these values by default, but I chose to set them explicitly. I have also tried without them, getting the same result.
And this is how I build my topology:
TopologyBuilder builder = new TopologyBuilder();
builder.setSpout(TWEETS_DATA_KAFKA_SPOUT_ID, kafkaSpout, kafkaSpoutParallelism)
.setNumTasks(kafkaSpoutNumberOfTasks);
builder.setBolt(ELASTICSEARCH_BOLT_ID, elasticsearchBolt, elasticsearchBoltParallelism)
.setNumTasks(elasticsearchBoltNumberOfTasks)
.shuffleGrouping(TWEETS_DATA_KAFKA_SPOUT_ID);
return builder.createTopology();
Before I run the topology locally I create the "twitter" index in Elasticsearch and a mapping "tweet" for this index.
This is what I get if I retrieve the mapping for my newly created type (curl -XGET 'http://localhost:9200/twitter/_mapping/tweet'):
{
"twitter": {
"mappings": {
"tweet": {
"properties": {
"text": {
"type": "string"
},
"tweetId": {
"type": "string"
}
}
}
}
}
}
I run the topology locally and this is what I get in my console when processing a tuple:
Processing received message FOR 6 TUPLE: source: tweets-data-kafka-spout:9, stream: default, id: {-8010897758788654352=-6240339405307942979}, [{"tweetId":"1","text":"hello"}]
Emitting: elasticsearch-bolt __ack_ack [-8010897758788654352 -6240339405307942979]
TRANSFERING tuple TASK: 2 TUPLE: source: elasticsearch-bolt:6, stream: __ack_ack, id: {}, [-8010897758788654352 -6240339405307942979]
BOLT ack TASK: 6 TIME: TUPLE: source: tweets-data-kafka-spout:9, stream: default, id: {-8010897758788654352=-6240339405307942979}, [{"tweetId":"1","text":"hello"}]
Execute done TUPLE source: tweets-data-kafka-spout:9, stream: default, id: {-8010897758788654352=-6240339405307942979}, [{"tweetId":"1","text":"hello"}] TASK: 6 DELTA:
So the tuples seems to be processed. However I don't have any document indexed in Elasticsearch.
I suppose I am doing something wrong when I set the configurations for EsBolt, maybe missing a configuration or something.
Documents will only be indexed once you reach the flush size, specified by es.storm.bolt.flush.entries.size
Alternately, you may set a TICK frequency that triggers a queue flush.
config.put(Config.TOPOLOGY_TICK_TUPLE_FREQ_SECS, 5);
By default, es-hadoop flushes on tick, as per the es.storm.bolt.tick.tuple.flush parameter.
I have also got the same issue, but when I looking for the es-Hadoop documents, I find because I was miss set the frequency that triggers a queue flush.Then I add a configurations to my store topology (es.storm.bolt.flush.entries.size ), it's fine.but when we setting the value for Config.TOPOLOGY_TICK_TUPLE_FREQ_SECS .it's throw an exception :java.lang.RuntimeException:java.lang.NullPointerException in bolt execute function. then we use debug mode to test my topology, I find the input tuple in bolt execute don't contain any entries, but this empty tuple is been triggered.
That's what I feel confusion. Don't the tuple will be emitted according to the setting time, Even though this tuple is empty after we set Config.TOPOLOGY_TICK_TUPLE_FREQ_SECS.i think which is a bug.
enter image description here
enter image description here
more information you can see:https://www.elastic.co/guide/en/elasticsearch/hadoop/current/storm.html