Commit message (Collapse) | Author | Age | ||
---|---|---|---|---|
... | ||||
* | Don't pass time to wait-condition-variable | Christopher Baines | 2023-12-02 | |
| | | | | | | | | | | My intent here was to avoid threads waiting indefinitely for a signal. Unfortunately, it seems that specifying a timeout can result in wait-condition-variable entering an endless loop internally (#67194). Hopefully avoid this by not passing a time to wait-condition-variable. Maybe some new threads/fibers will need adding that periodically signal these condition variables to replicate the timeout behaviour. | |||
* | Extend %file-port-name-canonicalization to hook threads | Christopher Baines | 2023-11-30 | |
| | ||||
* | Set %file-port-name-canonicalization to avoid readlink syscalls | Christopher Baines | 2023-11-28 | |
| | | | | As Guix does. | |||
* | Actually use a separate store connection | Christopher Baines | 2023-11-23 | |
| | | | | Fix the obvious bug in 3711cd8f1d901b424b2073dc54be4eec83c95482. | |||
* | Remove set-store-connection-timeout | Christopher Baines | 2023-11-20 | |
| | | | | As non blocking connections are being used now. | |||
* | Use a temporary store connection for substituting derivations | Christopher Baines | 2023-11-20 | |
| | ||||
* | Use port timeouts when checking substitutes | Christopher Baines | 2023-11-20 | |
| | | | | As I think this is better than the alarm syscall. | |||
* | Switch to (mostly) non blocking store connections in the agent | Christopher Baines | 2023-11-20 | |
| | | | | Pending #67245. | |||
* | Pass the store in to delete-outputs | Christopher Baines | 2023-11-20 | |
| | ||||
* | Pass the store in to get-output-details | Christopher Baines | 2023-11-20 | |
| | ||||
* | Don't shadow store in the agent | Christopher Baines | 2023-11-20 | |
| | ||||
* | Re-enable fiber preemption | Christopher Baines | 2023-11-18 | |
| | | | | As some fibers might be running for too long. | |||
* | Avoid readlink syscalls when reading derivations | Christopher Baines | 2023-11-17 | |
| | | | | | | | | | strace -c reports over 10,000 readlink syscalls when reading the derivation for the hello package. By just setting the %file-port-name-canonicalization fluid, this drops to less than 10. I'm not sure if this actually improves performance, but doing less is surely better. | |||
* | Track the metrics endpoint duration | Christopher Baines | 2023-11-17 | |
| | ||||
* | Call wait-condition-variable with longer timeouts | Christopher Baines | 2023-11-15 | |
| | | | | | | The hook processing threads seem to be getting stuck in wait-condition-variable where the timeout is in the past. Maybe setting a longer timeout will help avoid this. | |||
* | Remove some caching from reading derivations | Christopher Baines | 2023-11-11 | |
| | | | | As I'm not sure this is helpful. | |||
* | Try to avoid the metrics endpoint timing out | Christopher Baines | 2023-11-10 | |
| | | | | As this makes it harder to debug issues. | |||
* | Add a with-fibers-timeout utility | Christopher Baines | 2023-11-10 | |
| | ||||
* | Include the derivation_outputs in the data service event | Christopher Baines | 2023-10-16 | |
| | | | | If possible. | |||
* | Add new send-events-for-selected-builds procedure | Christopher Baines | 2023-10-16 | |
| | | | | Which is helpful when backfilling data in the data service. | |||
* | Use start-stack in make-worker-thread-channel | Christopher Baines | 2023-10-06 | |
| | | | | | | | | | | Not sure what this does, but it seems to work around Guile crashing in (backtrace) like: Backtrace: In guix-build-coordinator/utils/fibers.scm: Exception thrown while printing backtrace: In procedure vector-ref: Argument 2 out of range: 99 | |||
* | Don't output the whole log for substituting derivations | Christopher Baines | 2023-08-31 | |
| | | | | Just the last bit, as it could be very long. | |||
* | Exit when the server fails to start | Christopher Baines | 2023-08-29 | |
| | | | | To avoid the process half working. | |||
* | Replace request-builds-list by fold-builds | Christopher Baines | 2023-08-29 | |
| | | | | As this can wait until the stream is finished and properly close the port. | |||
* | Add more logging around hooks | Christopher Baines | 2023-08-29 | |
| | ||||
* | Fix some logging messages | Christopher Baines | 2023-08-29 | |
| | ||||
* | Stop closing the ports for json seq responses | Christopher Baines | 2023-08-29 | |
| | | | | As this breaks reading the stream. | |||
* | Properly close the port when making client requests | Christopher Baines | 2023-08-23 | |
| | ||||
* | Try and ensure that the worker threads don't disappear | Christopher Baines | 2023-08-22 | |
| | | | | Add some top level exception handling to guard against errors anywhere. | |||
* | Increase the open file limit | Christopher Baines | 2023-08-22 | |
| | | | | To hopefully make investigating why excessive ports/open files are being used. | |||
* | Use srfi-71 in the guix-data-service module | Christopher Baines | 2023-08-22 | |
| | ||||
* | Try and enable core dumps | Christopher Baines | 2023-08-20 | |
| | ||||
* | Pass the store in to substitute-derivation | Christopher Baines | 2023-08-20 | |
| | | | | To avoid an additional store connection. | |||
* | Use less database read threads | Christopher Baines | 2023-08-20 | |
| | | | | As I'm not sure that having so many read threads is helpful. | |||
* | Use less fibers threads and cooperative scheduling | Christopher Baines | 2023-08-20 | |
| | | | | Since I'm not sure preemption or more threads are useful. | |||
* | Include metric for file descriptors | Christopher Baines | 2023-08-20 | |
| | ||||
* | Switch to a more rigorous method for finding potential source refs | Christopher Baines | 2023-08-10 | |
| | ||||
* | Ignore self output references when looking for potential source refs | Christopher Baines | 2023-08-10 | |
| | ||||
* | Ensure references doesn't contain the empty string | Christopher Baines | 2023-08-10 | |
| | ||||
* | Simplify substitute-derivation | Christopher Baines | 2023-08-10 | |
| | | | | Assume ensure-path is available. | |||
* | Fix some log output | Christopher Baines | 2023-08-10 | |
| | ||||
* | Name more threads | Christopher Baines | 2023-08-09 | |
| | | | | To help with debugging. | |||
* | Don't use threads for substituting derivations | Christopher Baines | 2023-08-09 | |
| | | | | Instead use non-blocking store connections like Cuirass does. | |||
* | Name the fibers threads | Christopher Baines | 2023-08-09 | |
| | | | | To make debugging easier. | |||
* | Fix read-derivation getting the filename | Christopher Baines | 2023-08-03 | |
| | ||||
* | Keep less derivation files open when reading derivations | Christopher Baines | 2023-08-03 | |
| | | | | | This probably isn't the main problem with having too many open files, but it might help avoid bursts of open files. | |||
* | Remove crude fetching inputs timeout | Christopher Baines | 2023-08-02 | |
| | | | | As this breaks builds that rely on fetching large inputs. | |||
* | Reduce fibers parallelism | Christopher Baines | 2023-07-24 | |
| | | | | To reduce the number of threads, as lots is probably unnecessary. | |||
* | Make client requests mostly suspendable | Christopher Baines | 2023-07-24 | |
| | ||||
* | Add some logging to the build success publish hook | Christopher Baines | 2023-07-08 | |
| |