time: Location interprets wrong timezone (DST) with slim zoneinfo [1.15 backport] #42138
Comments
|
I believe this may be a broader issue. This commit to Edit: For those looking for a workaround here are a few options until a fix is released:
|
|
/cc @ianlancetaylor |
tklauser
added
the
NeedsInvestigation
|
When working on https://golang.org/cl/261363 (commit 5b509d9) I noticed that some of the tests in package I investigated this further and figured out that there is an issue with the way we cache location data, which lead to https://golang.org/cl/261877 (commit d83168e). In addition, this should probably haven been reported as an issue to make it easier to discover, which I didn't do. Sorry about that. I guess what we could do is backport the changes in |
|
Yes, I think this is likely due to the very recent change in the default tzdata install. We didn't notice until now because the change was so recent. Since fortunately @tklauser has already found and fixed the problem on tip, and since more and more people are going to be using slim tzdata installations, I agree that we should backport the fix. This should be a straightforward backport for 1.15. Unfortunately it is more work for 1.14, which doesn't have the I'll repurpose this issue for 1.15 and open a backport issue for 1.14. |
|
@gopherbot Please open backport for 1.14. Without a backport people using Go 1.14 with a new tzdata installation will not get the correct timezone information. |
|
Backport issue(s) opened: #42155 (for 1.14). Remember to create the cherry-pick CL(s) as soon as the patch is submitted to master, according to https://golang.org/wiki/MinorReleases. |
|
Change https://golang.org/cl/264301 mentions this issue: |
|
@hlubek It would be helpful if you could check that https://golang.org/cl/264301 fixes the problem for you. It is a patch against 1.15. Thanks. |
ianlancetaylor
added
the
CherryPickCandidate
Sure, will do! |
|
@hlubek I've tried using https://golang.org/cl/264301 and it didn't work as expected, unless I am missing something |
|
@dsantos I could reproduce that the patch fixes the issue reproduced in https://gist.github.com/hlubek/f46a73bc9d150cf1f2af585b0849e3d9 (Alpine with tzdata package) by using a patched Go toolchain to compile a binary for the image. Could you provider more details on your issue? |
|
Picking up from your example, this is how I've tested it: |
|
@dsantos Thanks, I can see it's failing and I digged a little deeper here. I took the test from https://go-review.googlesource.com/c/go/+/264301 and applied it to the golang image together with the patch and it failed. Before it was working for me. I think it's possible that the |
|
Change https://golang.org/cl/264939 mentions this issue: |
|
I implemented a fix for tip that could be applied to the backport: https://go-review.googlesource.com/c/go/+/264939 |
hlubek
changed the title
|
For anyone facing this issue, I switched to time zone data en Go using this: https://medium.com/@mhcbinder/using-local-time-in-a-golang-docker-container-built-from-scratch-2900af02fbaf. This caused a lot of problems in our site since last week, using America/Mexico_City. A new deploy with docker, would not calculate the correct tz data and we displayed different dates on our site. |
|
This caused us a few problems as we were using alpine:latest as our base image, which currently uses 2020c-r0 for tzdata. For anyone looking for a workaround, we pinned our version to alpine:3.8 and that seemed to do the trick (as it uses 2020a-r0). |
|
We all run into this on 17th Oct. oh, btw, we switched to distroless docker
image in order to solve the problem.
…On Tue, Oct 27, 2020 at 19:42 Rob Lowcock ***@***.***> wrote:
This caused us a few problems as we were using alpine:latest as our base
image, which currently uses 2020c-r0 for tzdata. For anyone looking for a
workaround, we pinned our version to alpine:3.8 and that seemed to do the
trick (as it uses 2020a-r0).
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#42138 (comment)>, or
unsubscribe
<https://github.com/notifications/unsubscribe-auth/AAE7EMDTHLRHDOUWVQDVWWTSM4A73ANCNFSM4S2YPIFA>
.
|
Pin to alpine:3.8.5 until golang/go#42138 is resolved, hopefully in go 1.15.4.
|
I am able to reproduce this with Alpine 3.12 but not Alpine 3.11. So you might not have to go back as far as 3.8. Can someone else confirm this? |
|
@magiconair I based the versions in my comment on the Alpine packages page which indicates that tzdata 2020c is current for all versions >3.8. I have just tested this (under docker) and can replicate the problem on versions 3.9-3.12 but 3.8 is OK. |
|
Approving per discussion in a release meeting. This is a serious issue without a workaround. This backport applies to both 1.15 (this issue) and 1.14 (#42155). |
dmitshur
added
CherryPickApproved
|
Closed by merging 414668c to release-branch.go1.15. |
|
Change https://golang.org/cl/266299 mentions this issue: |
|
gopherbot why are you closing this issue? |
Backport of part of https://golang.org/cl/261877 to support the slim tzdata format. As of tzdata 2020b, the default is to use the slim format. We need to support that format so that Go installations continue to work when tzdata is updated. Relevant part of the CL description: The reason for the failed tests was that when caching location data, the extended time format past the end of zone transitions was not considered. The respective change was introduced in (*Location).lookup by CL 215539. For #42138 Change-Id: I37f52a0917b2c6e3957e6b4612c8ef104c736e65 Reviewed-on: https://go-review.googlesource.com/c/go/+/264301 Trust: Ian Lance Taylor <[email protected]> Reviewed-by: Tobias Klauser <[email protected]>
|
Filed #42277 about gopherbot. |
|
@dmitshur @ianlancetaylor any idea when we can expect a release with this fix on 1.15? |
|
We usually do minor releases at the start of each month, so, if all goes well, next week. |
|
This tz issue has a considerable impact and the workaround (finding a compatible tzdata version) is not great, don't suppose there is any chance to have it earlier? |
|
Earlier than next week? It's already Thursday, and we don't want to make a release on Friday, so I don't see how we can do earlier than next week. There is an easy temporary workaround for Go 1.15: |
…data The extend information of a time zone file with last transition < now could result in a wrong cached zone because it used the zone of the last transition. This could lead to wrong zones in systems with slim zoneinfo. For #42216 Fixes #42138 Change-Id: I7c57c35b5cfa58482ac7925b5d86618c52f5444d Reviewed-on: https://go-review.googlesource.com/c/go/+/264939 Trust: Tobias Klauser <[email protected]> Run-TryBot: Tobias Klauser <[email protected]> TryBot-Result: Go Bot <[email protected]> Reviewed-by: Ian Lance Taylor <[email protected]> (cherry picked from commit 70e022e) Reviewed-on: https://go-review.googlesource.com/c/go/+/266299 Trust: Ian Lance Taylor <[email protected]> Run-TryBot: Ian Lance Taylor <[email protected]> Reviewed-by: Tobias Klauser <[email protected]>
|
This should now be fixed on the 1.15 release branch. |
curious: what's the downside of using UPDATE: relying on Go toolchain update schedule for time zone data updates isn't the best setup. e.g. there could be projects that for whatever reason need to be pegged at certain Go versions, and having that to mean the time zone info is also outdated isn't right 🙇♂️ Thanks Matt |
|
Sorry to bother but I'm able to understand if there is any workaround while the new version is released (which is not using the release branch one I mean). |
A temporary workaround is described in #42138 (comment) |
toothrot
removed
the
NeedsInvestigation
BREAKING CHANGES: * agent: The enable_central_service_config option now defaults to true. [GH-8746] * connect: Switch the default gateway port from 443 to 8443 to avoid assumption of Envoy running as root. [GH-9113] * connect: Update Envoy metrics names and labels for proxy listeners so that attributes like datacenter and namespace can be extracted. [GH-9207] * connect: intention destinations can no longer be reassigned [GH-8834] * raft: Raft protocol v2 is no longer supported. If currently using protocol v2 then an intermediate upgrade to a version supporting both v2 and v3 protocols will be necessary (1.0.0 - 1.8.x). Note that the Raft protocol configured with the raft_protocol setting and the Consul RPC protocol configured with the protocol setting and output by the consul version command are distinct and supported Consul RPC protocol versions are not altered. [GH-9103] * sentinel: (Consul Enterprise only) update to v0.16.0, which replaces whitelist and blacklist with allowlist and denylist * server: (Enterprise only) Pre-existing intentions defined with non-existent destination namespaces were non-functional and are erased during the upgrade process. This should not matter as these intentions had nothing to enforce. [GH-9186] * server: (OSS only) Pre-existing intentions defined with either a source or destination namespace value that is not "default" are rewritten or deleted during the upgrade process. Wildcards first attempt to downgrade to "default" unless an intention already exists, otherwise these non-functional intentions are deleted. [GH-9186] * xds: Drop support for Envoy versions 1.12.0, 1.12.1, 1.12.2, and 1.13.0, due to a lack of support for url_path in RBAC. [GH-8839] SECURITY: * Fix Consul Enterprise Namespace Config Entry Replication DoS. Previously an operator with service:write ACL permissions in a Consul Enterprise cluster could write a malicious config entry that caused infinite raft writes due to issues with the namespace replication logic. [CVE-2020-25201] [GH-9024] * Increase the permissions to read from the /connect/ca/configuration endpoint to operator:write. Previously Connect CA configuration, including the private key, set via this endpoint could be read back by an operator with operator:read privileges. CVE-2020-28053 [GH-9240] FEATURES: * agent: Add a new RPC endpoint for streaming cluster state change events to clients. * agent: Allow client agents to be configured with an advertised reconnect timeout to control how long until the nodes are reaped by others in the cluster. [GH-8781] * agent: moved ui config options to a new ui_config stanza in agent configuration and added new options to display service metrics in the UI. [GH-8694] * agent: return the default ACL policy to callers as a header [GH-9101] * autopilot: A new /v1/operator/autopilot/state HTTP API was created to give greater visibility into what autopilot is doing and how it has classified all the servers it is tracking. [GH-9103] * autopilot: Added a new consul operator autopilot state command to retrieve and view the Autopilot state from consul. [GH-9142] * cli: update snapshot inspect command to provide more detailed snapshot data [GH-8787] * connect: support defining intentions using layer 7 criteria [GH-8839] * telemetry: add initialization and definition for non-expiring key metrics in Prometheus [GH-9088] * telemetry: track node and service counts and emit them as metrics [GH-8603] * ui: If Prometheus is being used for monitoring the sidecars, the topology view can be configured to display overview metrics for the services. [GH-8858] * ui: Services using Connect with Envoy sidecars have a topology tab in the UI showing their upstream and downstream services. [GH-8788] * xds: use envoy's rbac filter to handle intentions entirely within envoy [GH-8569] IMPROVEMENTS: * agent: Return HTTP 429 when connections per clients limit (limits.http_max_conns_per_client) has been reached. [GH-8221] * agent: add path_allowlist config option to restrict metrics proxy queries [GH-9059] * agent: allow the /v1/connect/intentions/match endpoint to use the agent cache [GH-8875] * agent: protect the metrics proxy behind ACLs [GH-9099] * api: The v1/connect/ca/roots endpoint now accepts a pem=true query parameter and will return a PEM encoded certificate chain of all the certificates that would normally be in the JSON version of the response. [GH-8774] * api: support GetMeta() and GetNamespace() on all config entry kinds [GH-8764] * autopilot: (Enterprise Only) Autopilot now supports using both Redundancy Zones and Automated Upgrades together. [GH-9103] * checks: add health status to the failure message when gRPC healthchecks fail. [GH-8726] * chore: Update to Go 1.15 with mitigation for golang/go#42138 [GH-9036] * command: remove conditional envoy bootstrap generation for versions <=1.10.0 since those are not supported [GH-8855] * connect: The Vault provider will now automatically renew the lease of the token used, if supported. [GH-8560] * connect: add support for specifying load balancing policy in service-resolver [GH-8585] * connect: intentions are now managed as a new config entry kind "service-intentions" [GH-8834] * raft: Update raft to v1.2.0 to prevent non-voters from becoming eligible for leader elections and adding peer id as metric label to reduce cardinality in metric names [GH-8822] * server: (Consul Enterprise only) ensure that we also shutdown network segment serf instances on server shutdown [GH-8786] * server: break up Intention.Apply monolithic method [GH-9007] * server: create new memdb table for storing system metadata [GH-8703] * server: make sure that the various replication loggers use consistent logging [GH-8745] * server: remove config entry CAS in legacy intention API bridge code [GH-9151] * snapshot agent: Deregister critical snapshotting TTL check if leadership is transferred. * telemetry: All metrics should be present and available to prometheus scrapers when Consul starts. If any non-deprecated metrics are missing please submit an issue with its name. [GH-9198] * telemetry: add config flag telemetry { disable_compat_1.9 = (true|false) } to disable deprecated metrics in 1.9 [GH-8877] * telemetry: add counter consul.api.http with labels for each HTTP path and method. This is intended to replace consul.http... [GH-8877] * ui: Add the Upstreams and Exposed Paths tabs for services in mesh [GH-9141] * ui: Moves the Proxy health checks to be displayed with the Service health check under the Health Checks tab [GH-9141] * ui: Upstream and downstream services in the topology tab will show a visual indication if a deny intention or intention with L7 policies is configured. [GH-8846] * ui: add dashboard_url_template config option for external dashboard links [GH-9002] DEPRECATIONS: * Go 1.15 has dropped support for 32-bit binaries for Darwin, so darwin_386 builds will not be available for any 1.9.x+ releases. [GH-9036] * agent: ui, ui_dir and ui_content_path are now deprecated for use in agent configuration files. Use ui_config.{enable, dir, content_path} instead. The command arguments -ui, -ui-dir, and -ui-content-path remain supported. [GH-8694] * telemetry: The measurements in all of the consul.http... prefixed metrics have been migrated to consul.api.http. consul.http... prefixed metrics will be removed in a future version of Consul. [GH-8877] * telemetry: the disable_compat_1.9 config will cover more metrics deprecations in future 1.9 point releases. These metrics will be emitted twice for backwards compatibility - if the flag is true, only the new metric name will be written. [GH-9181] BUG FIXES: * agent: make the json/hcl decoding of ConnectProxyConfig fully work with CamelCase and snake_case [GH-8741] * agent: when enable_central_service_config is enabled ensure agent reload doesn't revert check state to critical [GH-8747] * api: Fixed a bug where the Check.GRPCUseTLS field could not be set using snake case. [GH-8771] * autopilot: (Enterprise Only) Previously servers in other zones would not be promoted when all servers in a second zone had failed. Now the actual behavior matches the docs and autopilot will promote a healthy non-voter from any zone to replace failure of an entire zone. [GH-9103] * autopilot: Prevent panic when requesting the autopilot health immediately after a leader is elected. [GH-9204] * command: when generating envoy bootstrap configs use the datacenter returned from the agent services endpoint [GH-9229] * connect: Fixed an issue where the Vault intermediate was not renewed in the primary datacenter. [GH-8784] * connect: fix Vault provider not respecting IntermediateCertTTL [GH-8646] * connect: fix connect sidecars registered via the API not being automatically deregistered with their parent service after an agent restart by persisting the LocallyRegisteredAsSidecar property. [GH-8924] * connect: use stronger validation that ingress gateways have compatible protocols defined for their upstreams [GH-8470] * license: (Enterprise only) Fixed an issue where the UI would see Namespaces and SSO as licensed when they were not. * license: (Enterprise only) Fixed an issue where warnings about Namespaces being unlicensed would be emitted erroneously. * namespace: (Enterprise Only) Fixed a bug that could case snapshot restoration to fail when it contained a namespace marked for deletion while still containing other resources in that namespace. [GH-9156] * namespace: (Enterprise Only) Fixed an issue where namespaced services and checks were not being deleted when the containing namespace was deleted. * raft: (Enterprise only) properly update consul server meta non_voter for non-voting Enterprise Consul servers [GH-8731] * server: skip deleted and deleting namespaces when migrating intentions to config entries [GH-9186] * telemetry: fixed a bug that caused logs to be flooded with [WARN] agent.router: Non-server in server-only area [GH-8685] * ui: show correct datacenter for gateways [GH-8704]
What version of Go are you using (
go version)?Does this issue reproduce with the latest release?
Yes, but it seems to be fixed in master with d83168e.
What operating system and processor architecture are you using (
go env)?go envOutputWhat did you do?
How to reproduce: https://gist.github.com/hlubek/f46a73bc9d150cf1f2af585b0849e3d9
tzdatapackageEurope/Berlin) uses wrong daylight saving information (e.g.CESTinstead ofCETfor dates after October, 25th in 2020)tzdatapackage in AlpinetzdataAlpine package is not installed and the embedded zoneinfotime/tzdatais usedWhat did you expect to see?
What did you see instead?
Notes
This is already fixed with d83168e but I wonder why there is no info about that issue (took me some time to figure it out) or a test for the code added toThe fix turns out to only work if the cached zone forzoneinfo_read.go. This is quite a standard setup for packaging Go programs so I could think many people are affected by this.now()is the same as the last transition. This broke after the last clock change on October, 25th for CET/CEST.The text was updated successfully, but these errors were encountered: