When notifications are sent to a large segment defined by a variety of parameters, the queries can quickly become complex and take minutes to execute because they may be returning millions of records from a set of tens of millions. On the read side, we support both transactional sending (ie, send to a particular subscriber at a particular engagement milestone) and sending to large audiences with specific characteristics (i.e.
POSTGRESQL SHARDING UPDATE
These records are written to very frequently - every time an app is opened, we will update the last time we saw that subscriber. We have over one billion monthly active subscriptions, and tens of billions of subscribers whose subscription status is unsubscribed. In terms of push notifications, a subscriber is identified as a push token, subscription status, and data tags (custom key : value pairs of string or number data that can be added to devices through our SDK) that support user segmentation. High-Level Overview of DataĪs a multichannel messaging platform, our primary data set is subscribers. Reference materials on the subjects we cover are provided as appropriate. One final note: our goal is to share our lessons learned at a high level rather than providing a detailed how-to guide. I recommend reading the High Level Overview of Data section first, then checking out the other sections listed below: There's a lot of information in this article - you can read it in order or jump around to different sections based on your interests. We're excited to share our lessons learned in this area because we had to figure out much of this the hard way, and we hope that sharing our experience will make things easier for others who are scaling PostgreSQL.
In this article, I’ll explain some of the challenges we’ve dealt with while scaling on PostgreSQL and the solutions we've put in place.
POSTGRESQL SHARDING UPGRADE
Our real-time segmentation features have benefited greatly from PostgreSQL’s performance, but we've also struggled at times due to bloat caused by our heavy write load and limitations of the PostgreSQL upgrade path. Over the years, we've scaled up to 75 terabyte (TB) of stored data across nearly 40 servers. For nearly a decade, the open-source relational database PostgreSQL has been a core part of OneSignal.
0 kommentar(er)
