Repository navigation
Conversation
…d a refused deploy popped, and record one node per IP The ephemeral-range warning now gives the other route: keep the range and reserve it with net.ipv4.ip_local_reserved_ports, persisted in sysctl.d. The worker reads its own network namespace, not the host's, so it cannot see the reservation; the warning says a box that reserved it can redeploy. A deploy refused because the running node's identity could not be read dropped a mismatch hold kept only in memory. It is restored now, like the stop and rename refusals already did. MystNodes allows one node per public IP; the catalog said unlimited.
|
Note Currently processing new changes in this PR. This may take a few minutes, please wait...
 |
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #679 +/- ##
==========================================
+ Coverage 93.96% 93.98% +0.01%
==========================================
Files 74 74
Lines 16882 16885 +3
==========================================
+ Hits 15864 15869 +5
+ Misses 1018 1016 -2
🚀 New features to boost your workflow:
|
…o only the in-memory pop on the refusal A reserved range keeps the kernel from handing its ports out, but a program that binds one by number still takes it. The refusal now puts back only the hold it popped and leaves the record on disk as it is, so a record written while the deploy ran is not overwritten.
…write replaces them
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Three loose ends from #677.
A box can keep its forwarded UDP range. The Fleet warning for a Mysterium node whose recorded range sits in Linux's ephemeral range (32768-60999) only offered moving the range and changing the router forward. Reserving the range in the host kernel works just as well and needs no router change. The warning now gives that route too: the exact
sysctl -w net.ipv4.ip_local_reserved_ports=<start>-<end>line, a note to add it to any ranges already listed there, and the line to put under/etc/sysctl.d/so it survives a reboot. The guide and UPGRADING step 3 say the same.The worker can't check the reservation. That sysctl is per network namespace, the worker runs on its own bridge, and the only thing CashPilot runs in the host's namespace is the fence helper, which runs in
onmode only. So the warning stays after you reserve the range, and it ends with "If you've reserved this range on the host, you can redeploy."A refused deploy dropped an in-memory hold.
deploy_rawpops the mismatch hold before it reads the running node's identity. If that read fails, the deploy is refused and the old node is left alone, but the hold stayed popped. When the on-disk MISMATCH record had failed to write, the memory copy was the only one, so the hold was gone. The refusal now calls_restore_identity(name, identity_before, held_before), as the stop and rename refusals already did;_roll_backrestores it on every create failure. I checked every other exit after the pop: a vanished old container goes on to create the replacement, and_roll_backcovers that path too.Mysterium allows one node per public IP. The guide said so, but the catalog had
devices_per_ip: 0(unlimited). The MystNodes help centre says "You can only run one MystNodes per public IP address": https://help.mystnodes.com/en/articles/3442543-multiple-nodes-on-a-single-ip-address. The catalog now says 1, and the source is listed in docs/research/per-ip-device-limits.md. What users see: a second Mysterium on the same machine, or on another machine behind the same public address, gets "will most likely earn nothing" in preflight instead of "earns less". Deploying stays allowed. The scaling limit was already 1 because the node has volumes, so only its reason text changes. The guide also stops saying the node runs with--net host.Tests: one for each part, plus seven mutants (dropping the restore, the reserve sentence, the recorded range, the dash form, the sysctl.d note, the "already reserved" line, and setting the value back to 0). Each one fails at least one test.
Summary by CodeRabbit