## New ElasticSearch storage option explanation in 9.2.0
Since v9.2.0, SkyWalking OAP provides 2 storage options for metrics/meter and records,
system environment variable is (`SW_STORAGE_ES_LOGIC_SHARDING`):
### No-Sharding Model (OAP default setting, `SW_STORAGE_ES_LOGIC_SHARDING = false`)
1. OAP merges all metrics/meter and records(without super datasets, such as segments) indices into one physical
index template `metrics-all` and `records-all`.
2. The logic index name would be present in columns `metric_table` and `record_table`.
3. If the logic column name has an alias (configured by `@ElasticSearch.Column()`), the alias would be the real physical column name.
### No-Sharding Model (`SW_STORAGE_ES_LOGIC_SHARDING = true `)
1. OAP shard metrics/meter indices into multi-physical indices as in the previous versions(one index template per metric/meter aggregation function).
2. Records and metrics without configuring aggregation function in `@MetricsFunction` and `@MeterFunction` would not be sharded.
3. The shard template name would be `metrics-aggregation function name` or `meter-aggregation function name` such as `metrics-count`,
and the logic index name would be present in column `metric_table`.
4. The OAP **would not** use the column alias, the logic column name would be the real physical column name.
**Notice**:
Users still could choose to adjust ElasticSearch's shard number(`SW_STORAGE_ES_INDEX_SHARDS_NUMBER`) to scale out no matter the option.
Move the forcedly flush out of each worker's #flush, which would only make the left data flushing into the storage ASAP. In my previous fix, it would increase the number of flushing as the number of workers, which is unnecessary too.
[Critical] Fix a low-performance issue of metrics persistent in the ElasticSearch storage implementation. One single metric could have to wait for an unnecessary 7~10s(System Env Variable SW_STORAGE_ES_FLUSH_INTERVAL).
This is a solid and critical performance issue, from the self-observability metric, we could find one metric that would need 7-10s to execute the data flush, which is not expected.
kezhenxu94
Fine0830
Kirs
Wan Kai
pg.yang
yswdqz
Tomasz Pytel
吴晟 Wu Sheng
mrproliu
Brandon Fergerson
StreamLang
xiao ru xing