You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Document SLO transform limitations related to misreported state and P… (#3123)
This PR adds a new limitation to the [Transforms
limitation](https://www.elastic.co/docs/explore-analyze/transforms/transform-limitations)
page about the behavior of SLO transforms at scale. When many SLO
transforms run concurrently, two issues can occur:
- Transforms may misreport their state and suggest deletion, even though
they remain valid.
- PIT requests may accumulate and create excess activity on the cluster,
leading to degraded performance.
The new section explains these limitations.
---------
Co-authored-by: István Zoltán Szabó <[email protected]>
Copy file name to clipboardExpand all lines: explore-analyze/transforms/transform-limitations.md
+49Lines changed: 49 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -123,6 +123,55 @@ If your data uses the [date nanosecond data type](elasticsearch://reference/elas
123
123
124
124
If you use ILM to have time-based indices, consider using the [Date index name](elasticsearch://reference/enrich-processor/date-index-name-processor.md) instead. The processor works without duplicated documents if your {{transform}} contains a `group_by` based on `date_histogram`.
125
125
126
+
### SLO transforms at scale may misreport state and increase PIT activity
127
+
128
+
When running a large number of SLO {{transforms}}, two types of limitations can appear:
129
+
130
+
- {{transforms-cap}} may [misreport](#transforms-inaccurate-errors) their state and suggest deletion even though they are still valid
131
+
- Point-in-time (PIT) requests may [accumulate](#transforms-pit-storms), creating extra activity on the cluster and slowing down operations
132
+
133
+
#### {{transforms-cap}} can return inaccurate errors that suggest deletion [transforms-inaccurate-errors]
134
+
135
+
The {{transforms-cap}} API and the {{transforms-cap}} page in {{kib}} (**Stack Management** > **{{transforms-cap}})** may display misleading error messages for {{transforms}} created by service level objectives (SLOs).
136
+
137
+
The message typically reads:
138
+
139
+
```console
140
+
Found task for transform [...], but no configuration for it. To delete this transform use DELETE with force=true.
141
+
```
142
+
Although the error suggests that the {{transform}} is broken and should be deleted, this is usually not the case. The error occurs because of a limitation in how the {{transforms}} are queried and presented by the API and UI. Deleting the {{transform}} can break related SLOs.
143
+
144
+
If you encounter this error, do not delete the {{transform}}. If a {{transform}} was already deleted, you can restore it by resetting the SLO, either in the {{kib}} or with the API:
145
+
146
+
- In {{kib}}, go to the **Observability** > **SLOs application**, open the menu for the affected SLO, and click **Reset**.
147
+
148
+
- If you use the API, run the following request:
149
+
150
+
```console
151
+
POST kbn:/api/observability/slos/<SLO_ID>/_reset
152
+
```
153
+
154
+
#### Large numbers of {{transform}} can cause PIT overloads [transforms-pit-overloads]
155
+
156
+
{{transforms-cap}} rely on point-in-time (PIT) searches to ensure that queries remain consistent during data changes. Each {{transform}} can open and close multiple PITs during its lifetime.
157
+
158
+
When many {{transforms}} run concurrently, especially in environments with large numbers of SLOs (hundreds to more than a thousand transforms), PITs can be opened and closed in quick succession. Because PITs are closed asynchronously, the close operation does not wait for the previous request to complete. This can create a backlog of PIT close requests, known as a PIT overload.
159
+
160
+
A PIT overload increases the memory pressure on hot nodes. This may result in rising memory usage and long garbage collection cycles. In practice, searches and writes may slow down or queue until the node recovers, and in severe cases a restart may be required.
161
+
162
+
As a workaround, you can disable PIT for a specific {{transform}}:
163
+
164
+
```console
165
+
POST _transform/<transform_id>/_update
166
+
{
167
+
"settings": {
168
+
"use_point_in_time": false
169
+
}
170
+
}
171
+
```
172
+
173
+
To fix this issue permanently, upgrade to the following versions or later: 8.18.8, 8.19.5, 9.0.8, 9.1.5, or 9.2.0.
174
+
126
175
## Limitations in {{kib}} [transform-ui-limitations]
127
176
128
177
### {{transforms-cap}} are visible in all {{kib}} spaces [transform-space-limitations]
0 commit comments