Details
-
Bug
-
Resolution: Incomplete
-
Major
-
None
-
Security Level: Public
-
None
-
4 node cluster on Ubuntu
-
2
Description
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Setup
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Bucket size : 3GB, Nodes : 4, keys inserted [64 - 256 bytes]
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Repro Steps
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Load high volume of data on the cluster[4 ndoes] , note - Data inserted > Low water Mark threshold.
2. Remove a server
3. Rebalance
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Error
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------
cbstats returns un-expectedly low resident memory ratio and high eject node ratio.
sample :
/opt/couchbase/bin/cbstats 10.1.3.67:11210 raw memory
ep_kv_size: 1616332569
ep_max_data_size: 3250585600
ep_mem_high_wat: 2437939200
ep_mem_low_wat: 1950351360
ep_oom_errors: 0
ep_overhead: 86679858
ep_tmp_oom_errors: 0
ep_value_size: 267558416
mem_used: 2421153021
tcmalloc_current_thread_cache_bytes: 7941064
tcmalloc_max_thread_cache_bytes: 33554432
tcmalloc_unmapped_bytes: 163840
total_allocated_bytes: 1662925672
total_fragmentation_bytes: 762578072
total_free_bytes: 153034752
total_heap_bytes: 2578538496
The diags from all the nodes are attached below and the cbstats monitored during the rebalance are attached as high_load.log
May -6 : Re-ran this test with the following setup, seeing drop in resident ratio( <50%) on one node and it has a higher fragmentation on it.
Ran w/ 1024 buckets, the resident ratio after rebalancing out one node, stays fairly ok( ~80-90%) on 2 nodes but drops sharply( <50% ) on the master node(10.1.3.92).
Build. - 1.8.1-802rel
Setup - 3GB bucketsize, 4 Nodes( 10.1.3.73, 10.1.3.70, 10.1.3.71, 10.1.3.92), Num of replicas = 2
vbuckets =1024, key-size=512-1k
mcosda: pytests/performance/mcsoda.py membase://10.1.3.92:8091 vbuckets=1024 doc-gen=0 doc-cache=0 ratio-creates=1 ratio-sets=1 min-value-size=512,1024 max-items=4000000 exit-after-creates=1 prefix=test_1_
- Load stopped before rebalance Out
- Total Items inserted = 2.28M
- Remove node and Rebalance Out a node.
- Smaller drop in resident ratio.
Diags attached as 10.1.3.70-8091-diag.txt.gz 10.1.3.72-8091-diag.txt.gz 10.1.3.92-8091-diag.txt.gz