Wikipedia:Replication lag
 | This is an information page. It is not an encyclopedic article, nor one of Wikipedia's policies or guidelines; rather, its purpose is to explain certain aspects of Wikipedia's norms, customs, technicalities, or practices. It may reflect differing levels of consensus and vetting. |
Replication lag is what happens when there are multiple data servers and one or more of these servers are delayed and out-of-sync with what would be considered the "current" state. The time between the "current" state and the state that the server is perceiving and returning is called the replication lag.
While this can apply to various servers and websites, at Wikipedia and Wikimedia it is generally not visible on production services. The servers with such a large delay are generally taken out of rotation until they are caught up again before most users will notice. In particular, the action API (api.php), which is widely used to get information from Wikimedia sites, is not affected by replication lag.
Toolforge and Cloud Services
[edit]Replication lag is an occasional problem for some bots and other tools that are hosted on Toolforge and Wikimedia Cloud Services. These tools will then no longer be able to get the latest information about conditions on the production services. They can only read the "past".
Replication lag occurs on these services because they make use of special copies of the actual wiki databases. These copies, also known as "the wiki replicas", filter out certain information that either should not be public or cannot be easily replicated (e.g. the text table). The services, while important, are also lower on the list of priorities than the main Wikimedia websites. Because of this, especially during maintenance operations to the databases, these copies sometimes become unavailable, out of sync or corrupted and need to be recovered. This can take anywhere from up to a couple of minutes to, in some cases, a day or more.
Effects of replication lag on Wikipedia
[edit]![[icon]](/wiki/File:Wiki_letter_w_cropped.svg){kind=small} | This section needs expansion. You can help by adding to it. |
Many bots refresh reports every few hours or more frequently. They run their queries on replicas of the live database, so if that replica is not being kept in sync with the live database, the reports will either fail to update or will display outdated information.
Examples of Wikipedia reports that are affected by replication lag
[edit] | This section needs expansion. You can help by adding to it. |
Most database reports listed at Wikipedia:Database reports are affected by replication lag. Some external tools, such as XTools and Quarry, are also affected.
Is there replag right now?
[edit]You can check this with the Replag tool. Under normal circumstances, all databases will show zero seconds (00:00:00) of replag.
Why do only some Wikimedia servers have this problem?
[edit]The databases are split up into slices (s1 through s8). A single Wikimedia website only uses 1 slice, but 1 slice generally provides service to many of the websites. The English Wikipedia is on s1; a full list of which website uses which slice is available on Toolforge. Slices are independent from each other and often won't be affected equally by replag.