Nikolay, > Do you know any real world usages of Ignite tracing? I know it was, but I cannot get a real use case right now because for a long time I haven't supported a long-lived Ignite 2 service. But I can bring a case to the table: I wanted to have the instrument when I work on performance:
1. Tones of similar operation have a latency lag for some period of time. 2. You cannot debug it because the issue happens after 5-10 minutes of loading. 3. Logging in is also useless because there are distributed operations. In this case tracing can trace one of 100K operations in the load stream. It shouldn't affect the load muck and show a correct trace. > Moreover, Ignite is an OLTP primary system and you can't trace any > specific operation, because in a regular cluster thousands or even > millions of operations run each second. If we talk about client scenarios, I see it like this: 1. The client starts a cluster with a trace that never traces any op. 2. If some measurements show that the load is stuck or latency dramatically increases for some ops. The client configured a filter for trace (by labe, by ration, by operation type, etc.). 3. After a day they analyze the captured traces. I do not see a reason why it is not possible for OLTP system. The main purpose is to minimize losses when the tracer is not tracing. On Thu, Jul 9, 2026 at 5:23 PM Nikolay Izhikov <[email protected]> wrote: > Hello, Vladislav. > > Thanks for the second opinion. > > Do you know any real world usages of Ignite tracing? > > > Tracing is a functionality that lets users evaluate in real time why an > operation takes longer. > > Don't think any user has such a question. > Moreover, Ignite is an OLTP primary system and you can't trace any > specific operation, because in a regular cluster thousands or even > millions of operations run each second. > > > чт, 9 июл. 2026 г. в 16:23, Vladislav Pyatkov <[email protected]>: > > > > Hi, > > > > Although all the points are true, the features (metrics, statistics) > provided in Ignite are not a replacement for tracing. Tracing is a > functionality that lets users evaluate in real time why an operation takes > longer. The current implementation may not be the best, but I am convinced > that the real reason is that it has not been maintained for a long time. > > What I mean is not that we should keep the OpenCensus API, but that we > need to plan how to implement new tracing, because there is no real > substitute for it in Ignite. > > > > Vladislav Pyatkov > > > > On Thu, Jul 9, 2026, 3:55 PM Mikhail Petrov <[email protected]> > wrote: > >> > >> Igniters, > >> > >> I would like to initiate a discussion on the deprecation and removal of > Ignite 2 Tracing. > >> > >> Motivation > >> > >> Ignite Tracing was introduced to provide distributed tracing > capabilities for Apache Ignite. However, over time several issues have > become apparent: > >> > >> 1. Ignite Tracing is based on OpenCensus, which has been deprecated and > is no longer maintained. [1] > >> 2. The implementation has never reached production readiness. It is > difficult to configure and, in some scenarios, introduces significant > performance overhead. > >> 3. The tracing implementation handles asynchronous Ignite operations > poorly, making the resulting traces difficult to interpret. > >> 4. It is tightly coupled with the Communication and Discovery > subsystems, increasing their complexity and making maintenance and > refactoring more difficult. > >> 5. Ignite Tracing has not received active maintenance for more than > three years. > >> > >> Before proceeding with removal, I would like to ask whether anyone is > aware of production deployments or real-world users who rely on Ignite > Tracing. > >> Such feedback would help us better assess the impact of its removal. > >> > >> > >> Alternatives > >> > >> Apache Ignite already provides several actively maintained mechanisms > for monitoring: > >> > >> 1. Ignite Performance Statistics [2] [3] > >> 2. Ignite Metric System [4] > >> 3. Ignite System Views [5] > >> > >> The aforementioned mechanisms are actively maintained and serve as the > primary means of monitoring Ignite. > >> > >> Regards, > >> Mikhail. > >> > >> [1] https://opentelemetry.io/blog/2023/sunsetting-opencensus > >> [2] > https://ignite.apache.org/docs/ignite2/latest/monitoring-metrics/performance-statistics > >> [3] > https://ignite.apache.org/docs/ignite2/latest/extensions-and-integrations/performance-statistics#print-statistics > >> [4] > https://ignite.apache.org/docs/ignite2/latest/monitoring-metrics/new-metrics-system > >> [5] > https://ignite.apache.org/docs/ignite2/latest/monitoring-metrics/system-views > >> > >> > >> > >> > -- Vladislav Pyatkov
